2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/badblocks.h>
38 #include <linux/sysctl.h>
39 #include <linux/seq_file.h>
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
57 #include "md-cluster.h"
60 static void autostart_arrays(int part);
63 /* pers_list is a list of registered personalities protected
65 * pers_lock does extra service to protect accesses to
66 * mddev->thread when the mutex cannot be held.
68 static LIST_HEAD(pers_list);
69 static DEFINE_SPINLOCK(pers_lock);
71 struct md_cluster_operations *md_cluster_ops;
72 EXPORT_SYMBOL(md_cluster_ops);
73 struct module *md_cluster_mod;
74 EXPORT_SYMBOL(md_cluster_mod);
76 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
77 static struct workqueue_struct *md_wq;
78 static struct workqueue_struct *md_misc_wq;
80 static int remove_and_add_spares(struct mddev *mddev,
81 struct md_rdev *this);
82 static void mddev_detach(struct mddev *mddev);
85 * Default number of read corrections we'll attempt on an rdev
86 * before ejecting it from the array. We divide the read error
87 * count by 2 for every hour elapsed between read errors.
89 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
91 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
92 * is 1000 KB/sec, so the extra system load does not show up that much.
93 * Increase it if you want to have more _guaranteed_ speed. Note that
94 * the RAID driver will use the maximum available bandwidth if the IO
95 * subsystem is idle. There is also an 'absolute maximum' reconstruction
96 * speed limit - in case reconstruction slows down your system despite
99 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
100 * or /sys/block/mdX/md/sync_speed_{min,max}
103 static int sysctl_speed_limit_min = 1000;
104 static int sysctl_speed_limit_max = 200000;
105 static inline int speed_min(struct mddev *mddev)
107 return mddev->sync_speed_min ?
108 mddev->sync_speed_min : sysctl_speed_limit_min;
111 static inline int speed_max(struct mddev *mddev)
113 return mddev->sync_speed_max ?
114 mddev->sync_speed_max : sysctl_speed_limit_max;
117 static struct ctl_table_header *raid_table_header;
119 static struct ctl_table raid_table[] = {
121 .procname = "speed_limit_min",
122 .data = &sysctl_speed_limit_min,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
128 .procname = "speed_limit_max",
129 .data = &sysctl_speed_limit_max,
130 .maxlen = sizeof(int),
131 .mode = S_IRUGO|S_IWUSR,
132 .proc_handler = proc_dointvec,
137 static struct ctl_table raid_dir_table[] = {
141 .mode = S_IRUGO|S_IXUGO,
147 static struct ctl_table raid_root_table[] = {
152 .child = raid_dir_table,
157 static const struct block_device_operations md_fops;
159 static int start_readonly;
162 * like bio_clone, but with a local bio set
165 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
170 if (!mddev || !mddev->bio_set)
171 return bio_alloc(gfp_mask, nr_iovecs);
173 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
178 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
180 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
183 if (!mddev || !mddev->bio_set)
184 return bio_clone(bio, gfp_mask);
186 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
188 EXPORT_SYMBOL_GPL(bio_clone_mddev);
191 * We have a system wide 'event count' that is incremented
192 * on any 'interesting' event, and readers of /proc/mdstat
193 * can use 'poll' or 'select' to find out when the event
197 * start array, stop array, error, add device, remove device,
198 * start build, activate spare
200 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
201 static atomic_t md_event_count;
202 void md_new_event(struct mddev *mddev)
204 atomic_inc(&md_event_count);
205 wake_up(&md_event_waiters);
207 EXPORT_SYMBOL_GPL(md_new_event);
210 * Enables to iterate over all existing md arrays
211 * all_mddevs_lock protects this list.
213 static LIST_HEAD(all_mddevs);
214 static DEFINE_SPINLOCK(all_mddevs_lock);
217 * iterates through all used mddevs in the system.
218 * We take care to grab the all_mddevs_lock whenever navigating
219 * the list, and to always hold a refcount when unlocked.
220 * Any code which breaks out of this loop while own
221 * a reference to the current mddev and must mddev_put it.
223 #define for_each_mddev(_mddev,_tmp) \
225 for (({ spin_lock(&all_mddevs_lock); \
226 _tmp = all_mddevs.next; \
228 ({ if (_tmp != &all_mddevs) \
229 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
230 spin_unlock(&all_mddevs_lock); \
231 if (_mddev) mddev_put(_mddev); \
232 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
233 _tmp != &all_mddevs;}); \
234 ({ spin_lock(&all_mddevs_lock); \
235 _tmp = _tmp->next;}) \
238 /* Rather than calling directly into the personality make_request function,
239 * IO requests come here first so that we can check if the device is
240 * being suspended pending a reconfiguration.
241 * We hold a refcount over the call to ->make_request. By the time that
242 * call has finished, the bio has been linked into some internal structure
243 * and so is visible to ->quiesce(), so we don't need the refcount any more.
245 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
247 const int rw = bio_data_dir(bio);
248 struct mddev *mddev = q->queuedata;
249 unsigned int sectors;
252 blk_queue_split(q, &bio, q->bio_split);
254 if (mddev == NULL || mddev->pers == NULL) {
256 return BLK_QC_T_NONE;
258 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
259 if (bio_sectors(bio) != 0)
260 bio->bi_error = -EROFS;
262 return BLK_QC_T_NONE;
264 smp_rmb(); /* Ensure implications of 'active' are visible */
266 if (mddev->suspended) {
269 prepare_to_wait(&mddev->sb_wait, &__wait,
270 TASK_UNINTERRUPTIBLE);
271 if (!mddev->suspended)
277 finish_wait(&mddev->sb_wait, &__wait);
279 atomic_inc(&mddev->active_io);
283 * save the sectors now since our bio can
284 * go away inside make_request
286 sectors = bio_sectors(bio);
287 /* bio could be mergeable after passing to underlayer */
288 bio->bi_opf &= ~REQ_NOMERGE;
289 mddev->pers->make_request(mddev, bio);
291 cpu = part_stat_lock();
292 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
293 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
296 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
297 wake_up(&mddev->sb_wait);
299 return BLK_QC_T_NONE;
302 /* mddev_suspend makes sure no new requests are submitted
303 * to the device, and that any requests that have been submitted
304 * are completely handled.
305 * Once mddev_detach() is called and completes, the module will be
308 void mddev_suspend(struct mddev *mddev)
310 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
311 if (mddev->suspended++)
314 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
315 mddev->pers->quiesce(mddev, 1);
317 del_timer_sync(&mddev->safemode_timer);
319 EXPORT_SYMBOL_GPL(mddev_suspend);
321 void mddev_resume(struct mddev *mddev)
323 if (--mddev->suspended)
325 wake_up(&mddev->sb_wait);
326 mddev->pers->quiesce(mddev, 0);
328 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
329 md_wakeup_thread(mddev->thread);
330 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
332 EXPORT_SYMBOL_GPL(mddev_resume);
334 int mddev_congested(struct mddev *mddev, int bits)
336 struct md_personality *pers = mddev->pers;
340 if (mddev->suspended)
342 else if (pers && pers->congested)
343 ret = pers->congested(mddev, bits);
347 EXPORT_SYMBOL_GPL(mddev_congested);
348 static int md_congested(void *data, int bits)
350 struct mddev *mddev = data;
351 return mddev_congested(mddev, bits);
355 * Generic flush handling for md
358 static void md_end_flush(struct bio *bio)
360 struct md_rdev *rdev = bio->bi_private;
361 struct mddev *mddev = rdev->mddev;
363 rdev_dec_pending(rdev, mddev);
365 if (atomic_dec_and_test(&mddev->flush_pending)) {
366 /* The pre-request flush has finished */
367 queue_work(md_wq, &mddev->flush_work);
372 static void md_submit_flush_data(struct work_struct *ws);
374 static void submit_flushes(struct work_struct *ws)
376 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
377 struct md_rdev *rdev;
379 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
380 atomic_set(&mddev->flush_pending, 1);
382 rdev_for_each_rcu(rdev, mddev)
383 if (rdev->raid_disk >= 0 &&
384 !test_bit(Faulty, &rdev->flags)) {
385 /* Take two references, one is dropped
386 * when request finishes, one after
387 * we reclaim rcu_read_lock
390 atomic_inc(&rdev->nr_pending);
391 atomic_inc(&rdev->nr_pending);
393 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
394 bi->bi_end_io = md_end_flush;
395 bi->bi_private = rdev;
396 bi->bi_bdev = rdev->bdev;
397 bio_set_op_attrs(bi, REQ_OP_WRITE, WRITE_FLUSH);
398 atomic_inc(&mddev->flush_pending);
401 rdev_dec_pending(rdev, mddev);
404 if (atomic_dec_and_test(&mddev->flush_pending))
405 queue_work(md_wq, &mddev->flush_work);
408 static void md_submit_flush_data(struct work_struct *ws)
410 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
411 struct bio *bio = mddev->flush_bio;
413 if (bio->bi_iter.bi_size == 0)
414 /* an empty barrier - all done */
417 bio->bi_opf &= ~REQ_PREFLUSH;
418 mddev->pers->make_request(mddev, bio);
421 mddev->flush_bio = NULL;
422 wake_up(&mddev->sb_wait);
425 void md_flush_request(struct mddev *mddev, struct bio *bio)
427 spin_lock_irq(&mddev->lock);
428 wait_event_lock_irq(mddev->sb_wait,
431 mddev->flush_bio = bio;
432 spin_unlock_irq(&mddev->lock);
434 INIT_WORK(&mddev->flush_work, submit_flushes);
435 queue_work(md_wq, &mddev->flush_work);
437 EXPORT_SYMBOL(md_flush_request);
439 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
441 struct mddev *mddev = cb->data;
442 md_wakeup_thread(mddev->thread);
445 EXPORT_SYMBOL(md_unplug);
447 static inline struct mddev *mddev_get(struct mddev *mddev)
449 atomic_inc(&mddev->active);
453 static void mddev_delayed_delete(struct work_struct *ws);
455 static void mddev_put(struct mddev *mddev)
457 struct bio_set *bs = NULL;
459 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
461 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
462 mddev->ctime == 0 && !mddev->hold_active) {
463 /* Array is not configured at all, and not held active,
465 list_del_init(&mddev->all_mddevs);
467 mddev->bio_set = NULL;
468 if (mddev->gendisk) {
469 /* We did a probe so need to clean up. Call
470 * queue_work inside the spinlock so that
471 * flush_workqueue() after mddev_find will
472 * succeed in waiting for the work to be done.
474 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
475 queue_work(md_misc_wq, &mddev->del_work);
479 spin_unlock(&all_mddevs_lock);
484 static void md_safemode_timeout(unsigned long data);
486 void mddev_init(struct mddev *mddev)
488 mutex_init(&mddev->open_mutex);
489 mutex_init(&mddev->reconfig_mutex);
490 mutex_init(&mddev->bitmap_info.mutex);
491 INIT_LIST_HEAD(&mddev->disks);
492 INIT_LIST_HEAD(&mddev->all_mddevs);
493 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
494 (unsigned long) mddev);
495 atomic_set(&mddev->active, 1);
496 atomic_set(&mddev->openers, 0);
497 atomic_set(&mddev->active_io, 0);
498 spin_lock_init(&mddev->lock);
499 atomic_set(&mddev->flush_pending, 0);
500 init_waitqueue_head(&mddev->sb_wait);
501 init_waitqueue_head(&mddev->recovery_wait);
502 mddev->reshape_position = MaxSector;
503 mddev->reshape_backwards = 0;
504 mddev->last_sync_action = "none";
505 mddev->resync_min = 0;
506 mddev->resync_max = MaxSector;
507 mddev->level = LEVEL_NONE;
509 EXPORT_SYMBOL_GPL(mddev_init);
511 static struct mddev *mddev_find_locked(dev_t unit)
515 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
516 if (mddev->unit == unit)
522 static struct mddev *mddev_find(dev_t unit)
524 struct mddev *mddev, *new = NULL;
526 if (unit && MAJOR(unit) != MD_MAJOR)
527 unit &= ~((1<<MdpMinorShift)-1);
530 spin_lock(&all_mddevs_lock);
533 mddev = mddev_find_locked(unit);
536 spin_unlock(&all_mddevs_lock);
542 list_add(&new->all_mddevs, &all_mddevs);
543 spin_unlock(&all_mddevs_lock);
544 new->hold_active = UNTIL_IOCTL;
548 /* find an unused unit number */
549 static int next_minor = 512;
550 int start = next_minor;
554 dev = MKDEV(MD_MAJOR, next_minor);
556 if (next_minor > MINORMASK)
558 if (next_minor == start) {
559 /* Oh dear, all in use. */
560 spin_unlock(&all_mddevs_lock);
565 is_free = !mddev_find_locked(dev);
568 new->md_minor = MINOR(dev);
569 new->hold_active = UNTIL_STOP;
570 list_add(&new->all_mddevs, &all_mddevs);
571 spin_unlock(&all_mddevs_lock);
574 spin_unlock(&all_mddevs_lock);
576 new = kzalloc(sizeof(*new), GFP_KERNEL);
581 if (MAJOR(unit) == MD_MAJOR)
582 new->md_minor = MINOR(unit);
584 new->md_minor = MINOR(unit) >> MdpMinorShift;
591 static struct attribute_group md_redundancy_group;
593 void mddev_unlock(struct mddev *mddev)
595 if (mddev->to_remove) {
596 /* These cannot be removed under reconfig_mutex as
597 * an access to the files will try to take reconfig_mutex
598 * while holding the file unremovable, which leads to
600 * So hold set sysfs_active while the remove in happeing,
601 * and anything else which might set ->to_remove or my
602 * otherwise change the sysfs namespace will fail with
603 * -EBUSY if sysfs_active is still set.
604 * We set sysfs_active under reconfig_mutex and elsewhere
605 * test it under the same mutex to ensure its correct value
608 struct attribute_group *to_remove = mddev->to_remove;
609 mddev->to_remove = NULL;
610 mddev->sysfs_active = 1;
611 mutex_unlock(&mddev->reconfig_mutex);
613 if (mddev->kobj.sd) {
614 if (to_remove != &md_redundancy_group)
615 sysfs_remove_group(&mddev->kobj, to_remove);
616 if (mddev->pers == NULL ||
617 mddev->pers->sync_request == NULL) {
618 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
619 if (mddev->sysfs_action)
620 sysfs_put(mddev->sysfs_action);
621 mddev->sysfs_action = NULL;
624 mddev->sysfs_active = 0;
626 mutex_unlock(&mddev->reconfig_mutex);
628 /* As we've dropped the mutex we need a spinlock to
629 * make sure the thread doesn't disappear
631 spin_lock(&pers_lock);
632 md_wakeup_thread(mddev->thread);
633 spin_unlock(&pers_lock);
635 EXPORT_SYMBOL_GPL(mddev_unlock);
637 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
639 struct md_rdev *rdev;
641 rdev_for_each_rcu(rdev, mddev)
642 if (rdev->desc_nr == nr)
647 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
649 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
651 struct md_rdev *rdev;
653 rdev_for_each(rdev, mddev)
654 if (rdev->bdev->bd_dev == dev)
660 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
662 struct md_rdev *rdev;
664 rdev_for_each_rcu(rdev, mddev)
665 if (rdev->bdev->bd_dev == dev)
671 static struct md_personality *find_pers(int level, char *clevel)
673 struct md_personality *pers;
674 list_for_each_entry(pers, &pers_list, list) {
675 if (level != LEVEL_NONE && pers->level == level)
677 if (strcmp(pers->name, clevel)==0)
683 /* return the offset of the super block in 512byte sectors */
684 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
686 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
687 return MD_NEW_SIZE_SECTORS(num_sectors);
690 static int alloc_disk_sb(struct md_rdev *rdev)
692 rdev->sb_page = alloc_page(GFP_KERNEL);
693 if (!rdev->sb_page) {
694 printk(KERN_ALERT "md: out of memory.\n");
701 void md_rdev_clear(struct md_rdev *rdev)
704 put_page(rdev->sb_page);
706 rdev->sb_page = NULL;
711 put_page(rdev->bb_page);
712 rdev->bb_page = NULL;
714 badblocks_exit(&rdev->badblocks);
716 EXPORT_SYMBOL_GPL(md_rdev_clear);
718 static void super_written(struct bio *bio)
720 struct md_rdev *rdev = bio->bi_private;
721 struct mddev *mddev = rdev->mddev;
724 printk("md: super_written gets error=%d\n", bio->bi_error);
725 md_error(mddev, rdev);
728 if (atomic_dec_and_test(&mddev->pending_writes))
729 wake_up(&mddev->sb_wait);
730 rdev_dec_pending(rdev, mddev);
734 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
735 sector_t sector, int size, struct page *page)
737 /* write first size bytes of page to sector of rdev
738 * Increment mddev->pending_writes before returning
739 * and decrement it on completion, waking up sb_wait
740 * if zero is reached.
741 * If an error occurred, call md_error
743 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
745 atomic_inc(&rdev->nr_pending);
747 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
748 bio->bi_iter.bi_sector = sector;
749 bio_add_page(bio, page, size, 0);
750 bio->bi_private = rdev;
751 bio->bi_end_io = super_written;
752 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH_FUA);
754 atomic_inc(&mddev->pending_writes);
758 void md_super_wait(struct mddev *mddev)
760 /* wait for all superblock writes that were scheduled to complete */
761 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
764 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
765 struct page *page, int op, int op_flags, bool metadata_op)
767 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
770 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
771 rdev->meta_bdev : rdev->bdev;
772 bio_set_op_attrs(bio, op, op_flags);
774 bio->bi_iter.bi_sector = sector + rdev->sb_start;
775 else if (rdev->mddev->reshape_position != MaxSector &&
776 (rdev->mddev->reshape_backwards ==
777 (sector >= rdev->mddev->reshape_position)))
778 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
780 bio->bi_iter.bi_sector = sector + rdev->data_offset;
781 bio_add_page(bio, page, size, 0);
783 submit_bio_wait(bio);
785 ret = !bio->bi_error;
789 EXPORT_SYMBOL_GPL(sync_page_io);
791 static int read_disk_sb(struct md_rdev *rdev, int size)
793 char b[BDEVNAME_SIZE];
798 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
804 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
805 bdevname(rdev->bdev,b));
809 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
811 return sb1->set_uuid0 == sb2->set_uuid0 &&
812 sb1->set_uuid1 == sb2->set_uuid1 &&
813 sb1->set_uuid2 == sb2->set_uuid2 &&
814 sb1->set_uuid3 == sb2->set_uuid3;
817 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
820 mdp_super_t *tmp1, *tmp2;
822 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
823 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
825 if (!tmp1 || !tmp2) {
827 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
835 * nr_disks is not constant
840 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
847 static u32 md_csum_fold(u32 csum)
849 csum = (csum & 0xffff) + (csum >> 16);
850 return (csum & 0xffff) + (csum >> 16);
853 static unsigned int calc_sb_csum(mdp_super_t *sb)
856 u32 *sb32 = (u32*)sb;
858 unsigned int disk_csum, csum;
860 disk_csum = sb->sb_csum;
863 for (i = 0; i < MD_SB_BYTES/4 ; i++)
865 csum = (newcsum & 0xffffffff) + (newcsum>>32);
868 /* This used to use csum_partial, which was wrong for several
869 * reasons including that different results are returned on
870 * different architectures. It isn't critical that we get exactly
871 * the same return value as before (we always csum_fold before
872 * testing, and that removes any differences). However as we
873 * know that csum_partial always returned a 16bit value on
874 * alphas, do a fold to maximise conformity to previous behaviour.
876 sb->sb_csum = md_csum_fold(disk_csum);
878 sb->sb_csum = disk_csum;
884 * Handle superblock details.
885 * We want to be able to handle multiple superblock formats
886 * so we have a common interface to them all, and an array of
887 * different handlers.
888 * We rely on user-space to write the initial superblock, and support
889 * reading and updating of superblocks.
890 * Interface methods are:
891 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
892 * loads and validates a superblock on dev.
893 * if refdev != NULL, compare superblocks on both devices
895 * 0 - dev has a superblock that is compatible with refdev
896 * 1 - dev has a superblock that is compatible and newer than refdev
897 * so dev should be used as the refdev in future
898 * -EINVAL superblock incompatible or invalid
899 * -othererror e.g. -EIO
901 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
902 * Verify that dev is acceptable into mddev.
903 * The first time, mddev->raid_disks will be 0, and data from
904 * dev should be merged in. Subsequent calls check that dev
905 * is new enough. Return 0 or -EINVAL
907 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
908 * Update the superblock for rdev with data in mddev
909 * This does not write to disc.
915 struct module *owner;
916 int (*load_super)(struct md_rdev *rdev,
917 struct md_rdev *refdev,
919 int (*validate_super)(struct mddev *mddev,
920 struct md_rdev *rdev);
921 void (*sync_super)(struct mddev *mddev,
922 struct md_rdev *rdev);
923 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
924 sector_t num_sectors);
925 int (*allow_new_offset)(struct md_rdev *rdev,
926 unsigned long long new_offset);
930 * Check that the given mddev has no bitmap.
932 * This function is called from the run method of all personalities that do not
933 * support bitmaps. It prints an error message and returns non-zero if mddev
934 * has a bitmap. Otherwise, it returns 0.
937 int md_check_no_bitmap(struct mddev *mddev)
939 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
941 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
942 mdname(mddev), mddev->pers->name);
945 EXPORT_SYMBOL(md_check_no_bitmap);
948 * load_super for 0.90.0
950 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
952 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
957 * Calculate the position of the superblock (512byte sectors),
958 * it's at the end of the disk.
960 * It also happens to be a multiple of 4Kb.
962 rdev->sb_start = calc_dev_sboffset(rdev);
964 ret = read_disk_sb(rdev, MD_SB_BYTES);
969 bdevname(rdev->bdev, b);
970 sb = page_address(rdev->sb_page);
972 if (sb->md_magic != MD_SB_MAGIC) {
973 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
978 if (sb->major_version != 0 ||
979 sb->minor_version < 90 ||
980 sb->minor_version > 91) {
981 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
982 sb->major_version, sb->minor_version,
987 if (sb->raid_disks <= 0)
990 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
991 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
996 rdev->preferred_minor = sb->md_minor;
997 rdev->data_offset = 0;
998 rdev->new_data_offset = 0;
999 rdev->sb_size = MD_SB_BYTES;
1000 rdev->badblocks.shift = -1;
1002 if (sb->level == LEVEL_MULTIPATH)
1005 rdev->desc_nr = sb->this_disk.number;
1011 mdp_super_t *refsb = page_address(refdev->sb_page);
1012 if (!uuid_equal(refsb, sb)) {
1013 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1014 b, bdevname(refdev->bdev,b2));
1017 if (!sb_equal(refsb, sb)) {
1018 printk(KERN_WARNING "md: %s has same UUID"
1019 " but different superblock to %s\n",
1020 b, bdevname(refdev->bdev, b2));
1024 ev2 = md_event(refsb);
1030 rdev->sectors = rdev->sb_start;
1031 /* Limit to 4TB as metadata cannot record more than that.
1032 * (not needed for Linear and RAID0 as metadata doesn't
1035 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1037 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1039 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1040 /* "this cannot possibly happen" ... */
1048 * validate_super for 0.90.0
1050 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1053 mdp_super_t *sb = page_address(rdev->sb_page);
1054 __u64 ev1 = md_event(sb);
1056 rdev->raid_disk = -1;
1057 clear_bit(Faulty, &rdev->flags);
1058 clear_bit(In_sync, &rdev->flags);
1059 clear_bit(Bitmap_sync, &rdev->flags);
1060 clear_bit(WriteMostly, &rdev->flags);
1062 if (mddev->raid_disks == 0) {
1063 mddev->major_version = 0;
1064 mddev->minor_version = sb->minor_version;
1065 mddev->patch_version = sb->patch_version;
1066 mddev->external = 0;
1067 mddev->chunk_sectors = sb->chunk_size >> 9;
1068 mddev->ctime = sb->ctime;
1069 mddev->utime = sb->utime;
1070 mddev->level = sb->level;
1071 mddev->clevel[0] = 0;
1072 mddev->layout = sb->layout;
1073 mddev->raid_disks = sb->raid_disks;
1074 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1075 mddev->events = ev1;
1076 mddev->bitmap_info.offset = 0;
1077 mddev->bitmap_info.space = 0;
1078 /* bitmap can use 60 K after the 4K superblocks */
1079 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1080 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1081 mddev->reshape_backwards = 0;
1083 if (mddev->minor_version >= 91) {
1084 mddev->reshape_position = sb->reshape_position;
1085 mddev->delta_disks = sb->delta_disks;
1086 mddev->new_level = sb->new_level;
1087 mddev->new_layout = sb->new_layout;
1088 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1089 if (mddev->delta_disks < 0)
1090 mddev->reshape_backwards = 1;
1092 mddev->reshape_position = MaxSector;
1093 mddev->delta_disks = 0;
1094 mddev->new_level = mddev->level;
1095 mddev->new_layout = mddev->layout;
1096 mddev->new_chunk_sectors = mddev->chunk_sectors;
1099 if (sb->state & (1<<MD_SB_CLEAN))
1100 mddev->recovery_cp = MaxSector;
1102 if (sb->events_hi == sb->cp_events_hi &&
1103 sb->events_lo == sb->cp_events_lo) {
1104 mddev->recovery_cp = sb->recovery_cp;
1106 mddev->recovery_cp = 0;
1109 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1110 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1111 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1112 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1114 mddev->max_disks = MD_SB_DISKS;
1116 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1117 mddev->bitmap_info.file == NULL) {
1118 mddev->bitmap_info.offset =
1119 mddev->bitmap_info.default_offset;
1120 mddev->bitmap_info.space =
1121 mddev->bitmap_info.default_space;
1124 } else if (mddev->pers == NULL) {
1125 /* Insist on good event counter while assembling, except
1126 * for spares (which don't need an event count) */
1128 if (sb->disks[rdev->desc_nr].state & (
1129 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1130 if (ev1 < mddev->events)
1132 } else if (mddev->bitmap) {
1133 /* if adding to array with a bitmap, then we can accept an
1134 * older device ... but not too old.
1136 if (ev1 < mddev->bitmap->events_cleared)
1138 if (ev1 < mddev->events)
1139 set_bit(Bitmap_sync, &rdev->flags);
1141 if (ev1 < mddev->events)
1142 /* just a hot-add of a new device, leave raid_disk at -1 */
1146 if (mddev->level != LEVEL_MULTIPATH) {
1147 desc = sb->disks + rdev->desc_nr;
1149 if (desc->state & (1<<MD_DISK_FAULTY))
1150 set_bit(Faulty, &rdev->flags);
1151 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1152 desc->raid_disk < mddev->raid_disks */) {
1153 set_bit(In_sync, &rdev->flags);
1154 rdev->raid_disk = desc->raid_disk;
1155 rdev->saved_raid_disk = desc->raid_disk;
1156 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1157 /* active but not in sync implies recovery up to
1158 * reshape position. We don't know exactly where
1159 * that is, so set to zero for now */
1160 if (mddev->minor_version >= 91) {
1161 rdev->recovery_offset = 0;
1162 rdev->raid_disk = desc->raid_disk;
1165 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1166 set_bit(WriteMostly, &rdev->flags);
1167 } else /* MULTIPATH are always insync */
1168 set_bit(In_sync, &rdev->flags);
1173 * sync_super for 0.90.0
1175 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1178 struct md_rdev *rdev2;
1179 int next_spare = mddev->raid_disks;
1181 /* make rdev->sb match mddev data..
1184 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1185 * 3/ any empty disks < next_spare become removed
1187 * disks[0] gets initialised to REMOVED because
1188 * we cannot be sure from other fields if it has
1189 * been initialised or not.
1192 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1194 rdev->sb_size = MD_SB_BYTES;
1196 sb = page_address(rdev->sb_page);
1198 memset(sb, 0, sizeof(*sb));
1200 sb->md_magic = MD_SB_MAGIC;
1201 sb->major_version = mddev->major_version;
1202 sb->patch_version = mddev->patch_version;
1203 sb->gvalid_words = 0; /* ignored */
1204 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1205 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1206 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1207 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1209 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1210 sb->level = mddev->level;
1211 sb->size = mddev->dev_sectors / 2;
1212 sb->raid_disks = mddev->raid_disks;
1213 sb->md_minor = mddev->md_minor;
1214 sb->not_persistent = 0;
1215 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1217 sb->events_hi = (mddev->events>>32);
1218 sb->events_lo = (u32)mddev->events;
1220 if (mddev->reshape_position == MaxSector)
1221 sb->minor_version = 90;
1223 sb->minor_version = 91;
1224 sb->reshape_position = mddev->reshape_position;
1225 sb->new_level = mddev->new_level;
1226 sb->delta_disks = mddev->delta_disks;
1227 sb->new_layout = mddev->new_layout;
1228 sb->new_chunk = mddev->new_chunk_sectors << 9;
1230 mddev->minor_version = sb->minor_version;
1233 sb->recovery_cp = mddev->recovery_cp;
1234 sb->cp_events_hi = (mddev->events>>32);
1235 sb->cp_events_lo = (u32)mddev->events;
1236 if (mddev->recovery_cp == MaxSector)
1237 sb->state = (1<< MD_SB_CLEAN);
1239 sb->recovery_cp = 0;
1241 sb->layout = mddev->layout;
1242 sb->chunk_size = mddev->chunk_sectors << 9;
1244 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1245 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1247 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1248 rdev_for_each(rdev2, mddev) {
1251 int is_active = test_bit(In_sync, &rdev2->flags);
1253 if (rdev2->raid_disk >= 0 &&
1254 sb->minor_version >= 91)
1255 /* we have nowhere to store the recovery_offset,
1256 * but if it is not below the reshape_position,
1257 * we can piggy-back on that.
1260 if (rdev2->raid_disk < 0 ||
1261 test_bit(Faulty, &rdev2->flags))
1264 desc_nr = rdev2->raid_disk;
1266 desc_nr = next_spare++;
1267 rdev2->desc_nr = desc_nr;
1268 d = &sb->disks[rdev2->desc_nr];
1270 d->number = rdev2->desc_nr;
1271 d->major = MAJOR(rdev2->bdev->bd_dev);
1272 d->minor = MINOR(rdev2->bdev->bd_dev);
1274 d->raid_disk = rdev2->raid_disk;
1276 d->raid_disk = rdev2->desc_nr; /* compatibility */
1277 if (test_bit(Faulty, &rdev2->flags))
1278 d->state = (1<<MD_DISK_FAULTY);
1279 else if (is_active) {
1280 d->state = (1<<MD_DISK_ACTIVE);
1281 if (test_bit(In_sync, &rdev2->flags))
1282 d->state |= (1<<MD_DISK_SYNC);
1290 if (test_bit(WriteMostly, &rdev2->flags))
1291 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1293 /* now set the "removed" and "faulty" bits on any missing devices */
1294 for (i=0 ; i < mddev->raid_disks ; i++) {
1295 mdp_disk_t *d = &sb->disks[i];
1296 if (d->state == 0 && d->number == 0) {
1299 d->state = (1<<MD_DISK_REMOVED);
1300 d->state |= (1<<MD_DISK_FAULTY);
1304 sb->nr_disks = nr_disks;
1305 sb->active_disks = active;
1306 sb->working_disks = working;
1307 sb->failed_disks = failed;
1308 sb->spare_disks = spare;
1310 sb->this_disk = sb->disks[rdev->desc_nr];
1311 sb->sb_csum = calc_sb_csum(sb);
1315 * rdev_size_change for 0.90.0
1317 static unsigned long long
1318 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1320 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1321 return 0; /* component must fit device */
1322 if (rdev->mddev->bitmap_info.offset)
1323 return 0; /* can't move bitmap */
1324 rdev->sb_start = calc_dev_sboffset(rdev);
1325 if (!num_sectors || num_sectors > rdev->sb_start)
1326 num_sectors = rdev->sb_start;
1327 /* Limit to 4TB as metadata cannot record more than that.
1328 * 4TB == 2^32 KB, or 2*2^32 sectors.
1330 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1331 rdev->mddev->level >= 1)
1332 num_sectors = (sector_t)(2ULL << 32) - 2;
1333 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1335 md_super_wait(rdev->mddev);
1340 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1342 /* non-zero offset changes not possible with v0.90 */
1343 return new_offset == 0;
1347 * version 1 superblock
1350 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1354 unsigned long long newcsum;
1355 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1356 __le32 *isuper = (__le32*)sb;
1358 disk_csum = sb->sb_csum;
1361 for (; size >= 4; size -= 4)
1362 newcsum += le32_to_cpu(*isuper++);
1365 newcsum += le16_to_cpu(*(__le16*) isuper);
1367 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1368 sb->sb_csum = disk_csum;
1369 return cpu_to_le32(csum);
1372 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1374 struct mdp_superblock_1 *sb;
1378 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1382 * Calculate the position of the superblock in 512byte sectors.
1383 * It is always aligned to a 4K boundary and
1384 * depeding on minor_version, it can be:
1385 * 0: At least 8K, but less than 12K, from end of device
1386 * 1: At start of device
1387 * 2: 4K from start of device.
1389 switch(minor_version) {
1391 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1393 sb_start &= ~(sector_t)(4*2-1);
1404 rdev->sb_start = sb_start;
1406 /* superblock is rarely larger than 1K, but it can be larger,
1407 * and it is safe to read 4k, so we do that
1409 ret = read_disk_sb(rdev, 4096);
1410 if (ret) return ret;
1412 sb = page_address(rdev->sb_page);
1414 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1415 sb->major_version != cpu_to_le32(1) ||
1416 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1417 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1418 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1421 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1422 printk("md: invalid superblock checksum on %s\n",
1423 bdevname(rdev->bdev,b));
1426 if (le64_to_cpu(sb->data_size) < 10) {
1427 printk("md: data_size too small on %s\n",
1428 bdevname(rdev->bdev,b));
1433 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1434 /* Some padding is non-zero, might be a new feature */
1437 rdev->preferred_minor = 0xffff;
1438 rdev->data_offset = le64_to_cpu(sb->data_offset);
1439 rdev->new_data_offset = rdev->data_offset;
1440 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1441 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1442 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1443 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1445 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1446 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1447 if (rdev->sb_size & bmask)
1448 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1451 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1454 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1457 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1460 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1462 if (!rdev->bb_page) {
1463 rdev->bb_page = alloc_page(GFP_KERNEL);
1467 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1468 rdev->badblocks.count == 0) {
1469 /* need to load the bad block list.
1470 * Currently we limit it to one page.
1476 int sectors = le16_to_cpu(sb->bblog_size);
1477 if (sectors > (PAGE_SIZE / 512))
1479 offset = le32_to_cpu(sb->bblog_offset);
1482 bb_sector = (long long)offset;
1483 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1484 rdev->bb_page, REQ_OP_READ, 0, true))
1486 bbp = (u64 *)page_address(rdev->bb_page);
1487 rdev->badblocks.shift = sb->bblog_shift;
1488 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1489 u64 bb = le64_to_cpu(*bbp);
1490 int count = bb & (0x3ff);
1491 u64 sector = bb >> 10;
1492 sector <<= sb->bblog_shift;
1493 count <<= sb->bblog_shift;
1496 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1499 } else if (sb->bblog_offset != 0)
1500 rdev->badblocks.shift = 0;
1506 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1508 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1509 sb->level != refsb->level ||
1510 sb->layout != refsb->layout ||
1511 sb->chunksize != refsb->chunksize) {
1512 printk(KERN_WARNING "md: %s has strangely different"
1513 " superblock to %s\n",
1514 bdevname(rdev->bdev,b),
1515 bdevname(refdev->bdev,b2));
1518 ev1 = le64_to_cpu(sb->events);
1519 ev2 = le64_to_cpu(refsb->events);
1526 if (minor_version) {
1527 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1528 sectors -= rdev->data_offset;
1530 sectors = rdev->sb_start;
1531 if (sectors < le64_to_cpu(sb->data_size))
1533 rdev->sectors = le64_to_cpu(sb->data_size);
1537 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1539 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1540 __u64 ev1 = le64_to_cpu(sb->events);
1542 rdev->raid_disk = -1;
1543 clear_bit(Faulty, &rdev->flags);
1544 clear_bit(In_sync, &rdev->flags);
1545 clear_bit(Bitmap_sync, &rdev->flags);
1546 clear_bit(WriteMostly, &rdev->flags);
1548 if (mddev->raid_disks == 0) {
1549 mddev->major_version = 1;
1550 mddev->patch_version = 0;
1551 mddev->external = 0;
1552 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1553 mddev->ctime = le64_to_cpu(sb->ctime);
1554 mddev->utime = le64_to_cpu(sb->utime);
1555 mddev->level = le32_to_cpu(sb->level);
1556 mddev->clevel[0] = 0;
1557 mddev->layout = le32_to_cpu(sb->layout);
1558 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1559 mddev->dev_sectors = le64_to_cpu(sb->size);
1560 mddev->events = ev1;
1561 mddev->bitmap_info.offset = 0;
1562 mddev->bitmap_info.space = 0;
1563 /* Default location for bitmap is 1K after superblock
1564 * using 3K - total of 4K
1566 mddev->bitmap_info.default_offset = 1024 >> 9;
1567 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1568 mddev->reshape_backwards = 0;
1570 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1571 memcpy(mddev->uuid, sb->set_uuid, 16);
1573 mddev->max_disks = (4096-256)/2;
1575 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1576 mddev->bitmap_info.file == NULL) {
1577 mddev->bitmap_info.offset =
1578 (__s32)le32_to_cpu(sb->bitmap_offset);
1579 /* Metadata doesn't record how much space is available.
1580 * For 1.0, we assume we can use up to the superblock
1581 * if before, else to 4K beyond superblock.
1582 * For others, assume no change is possible.
1584 if (mddev->minor_version > 0)
1585 mddev->bitmap_info.space = 0;
1586 else if (mddev->bitmap_info.offset > 0)
1587 mddev->bitmap_info.space =
1588 8 - mddev->bitmap_info.offset;
1590 mddev->bitmap_info.space =
1591 -mddev->bitmap_info.offset;
1594 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1595 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1596 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1597 mddev->new_level = le32_to_cpu(sb->new_level);
1598 mddev->new_layout = le32_to_cpu(sb->new_layout);
1599 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1600 if (mddev->delta_disks < 0 ||
1601 (mddev->delta_disks == 0 &&
1602 (le32_to_cpu(sb->feature_map)
1603 & MD_FEATURE_RESHAPE_BACKWARDS)))
1604 mddev->reshape_backwards = 1;
1606 mddev->reshape_position = MaxSector;
1607 mddev->delta_disks = 0;
1608 mddev->new_level = mddev->level;
1609 mddev->new_layout = mddev->layout;
1610 mddev->new_chunk_sectors = mddev->chunk_sectors;
1613 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1614 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1615 } else if (mddev->pers == NULL) {
1616 /* Insist of good event counter while assembling, except for
1617 * spares (which don't need an event count) */
1619 if (rdev->desc_nr >= 0 &&
1620 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1621 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1622 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1623 if (ev1 < mddev->events)
1625 } else if (mddev->bitmap) {
1626 /* If adding to array with a bitmap, then we can accept an
1627 * older device, but not too old.
1629 if (ev1 < mddev->bitmap->events_cleared)
1631 if (ev1 < mddev->events)
1632 set_bit(Bitmap_sync, &rdev->flags);
1634 if (ev1 < mddev->events)
1635 /* just a hot-add of a new device, leave raid_disk at -1 */
1638 if (mddev->level != LEVEL_MULTIPATH) {
1640 if (rdev->desc_nr < 0 ||
1641 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1642 role = MD_DISK_ROLE_SPARE;
1645 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1647 case MD_DISK_ROLE_SPARE: /* spare */
1649 case MD_DISK_ROLE_FAULTY: /* faulty */
1650 set_bit(Faulty, &rdev->flags);
1652 case MD_DISK_ROLE_JOURNAL: /* journal device */
1653 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1654 /* journal device without journal feature */
1656 "md: journal device provided without journal feature, ignoring the device\n");
1659 set_bit(Journal, &rdev->flags);
1660 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1661 rdev->raid_disk = 0;
1664 rdev->saved_raid_disk = role;
1665 if ((le32_to_cpu(sb->feature_map) &
1666 MD_FEATURE_RECOVERY_OFFSET)) {
1667 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1668 if (!(le32_to_cpu(sb->feature_map) &
1669 MD_FEATURE_RECOVERY_BITMAP))
1670 rdev->saved_raid_disk = -1;
1673 * If the array is FROZEN, then the device can't
1674 * be in_sync with rest of array.
1676 if (!test_bit(MD_RECOVERY_FROZEN,
1678 set_bit(In_sync, &rdev->flags);
1680 rdev->raid_disk = role;
1683 if (sb->devflags & WriteMostly1)
1684 set_bit(WriteMostly, &rdev->flags);
1685 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1686 set_bit(Replacement, &rdev->flags);
1687 } else /* MULTIPATH are always insync */
1688 set_bit(In_sync, &rdev->flags);
1693 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1695 struct mdp_superblock_1 *sb;
1696 struct md_rdev *rdev2;
1698 /* make rdev->sb match mddev and rdev data. */
1700 sb = page_address(rdev->sb_page);
1702 sb->feature_map = 0;
1704 sb->recovery_offset = cpu_to_le64(0);
1705 memset(sb->pad3, 0, sizeof(sb->pad3));
1707 sb->utime = cpu_to_le64((__u64)mddev->utime);
1708 sb->events = cpu_to_le64(mddev->events);
1710 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1711 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1712 sb->resync_offset = cpu_to_le64(MaxSector);
1714 sb->resync_offset = cpu_to_le64(0);
1716 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1718 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1719 sb->size = cpu_to_le64(mddev->dev_sectors);
1720 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1721 sb->level = cpu_to_le32(mddev->level);
1722 sb->layout = cpu_to_le32(mddev->layout);
1724 if (test_bit(WriteMostly, &rdev->flags))
1725 sb->devflags |= WriteMostly1;
1727 sb->devflags &= ~WriteMostly1;
1728 sb->data_offset = cpu_to_le64(rdev->data_offset);
1729 sb->data_size = cpu_to_le64(rdev->sectors);
1731 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1732 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1733 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1736 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1737 !test_bit(In_sync, &rdev->flags)) {
1739 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1740 sb->recovery_offset =
1741 cpu_to_le64(rdev->recovery_offset);
1742 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1744 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1746 /* Note: recovery_offset and journal_tail share space */
1747 if (test_bit(Journal, &rdev->flags))
1748 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1749 if (test_bit(Replacement, &rdev->flags))
1751 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1753 if (mddev->reshape_position != MaxSector) {
1754 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1755 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1756 sb->new_layout = cpu_to_le32(mddev->new_layout);
1757 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1758 sb->new_level = cpu_to_le32(mddev->new_level);
1759 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1760 if (mddev->delta_disks == 0 &&
1761 mddev->reshape_backwards)
1763 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1764 if (rdev->new_data_offset != rdev->data_offset) {
1766 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1767 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1768 - rdev->data_offset));
1772 if (mddev_is_clustered(mddev))
1773 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1775 if (rdev->badblocks.count == 0)
1776 /* Nothing to do for bad blocks*/ ;
1777 else if (sb->bblog_offset == 0)
1778 /* Cannot record bad blocks on this device */
1779 md_error(mddev, rdev);
1781 struct badblocks *bb = &rdev->badblocks;
1782 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1784 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1789 seq = read_seqbegin(&bb->lock);
1791 memset(bbp, 0xff, PAGE_SIZE);
1793 for (i = 0 ; i < bb->count ; i++) {
1794 u64 internal_bb = p[i];
1795 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1796 | BB_LEN(internal_bb));
1797 bbp[i] = cpu_to_le64(store_bb);
1800 if (read_seqretry(&bb->lock, seq))
1803 bb->sector = (rdev->sb_start +
1804 (int)le32_to_cpu(sb->bblog_offset));
1805 bb->size = le16_to_cpu(sb->bblog_size);
1810 rdev_for_each(rdev2, mddev)
1811 if (rdev2->desc_nr+1 > max_dev)
1812 max_dev = rdev2->desc_nr+1;
1814 if (max_dev > le32_to_cpu(sb->max_dev)) {
1816 sb->max_dev = cpu_to_le32(max_dev);
1817 rdev->sb_size = max_dev * 2 + 256;
1818 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1819 if (rdev->sb_size & bmask)
1820 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1822 max_dev = le32_to_cpu(sb->max_dev);
1824 for (i=0; i<max_dev;i++)
1825 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1827 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1828 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1830 rdev_for_each(rdev2, mddev) {
1832 if (test_bit(Faulty, &rdev2->flags))
1833 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1834 else if (test_bit(In_sync, &rdev2->flags))
1835 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1836 else if (test_bit(Journal, &rdev2->flags))
1837 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1838 else if (rdev2->raid_disk >= 0)
1839 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1841 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1844 sb->sb_csum = calc_sb_1_csum(sb);
1847 static unsigned long long
1848 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1850 struct mdp_superblock_1 *sb;
1851 sector_t max_sectors;
1852 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1853 return 0; /* component must fit device */
1854 if (rdev->data_offset != rdev->new_data_offset)
1855 return 0; /* too confusing */
1856 if (rdev->sb_start < rdev->data_offset) {
1857 /* minor versions 1 and 2; superblock before data */
1858 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1859 max_sectors -= rdev->data_offset;
1860 if (!num_sectors || num_sectors > max_sectors)
1861 num_sectors = max_sectors;
1862 } else if (rdev->mddev->bitmap_info.offset) {
1863 /* minor version 0 with bitmap we can't move */
1866 /* minor version 0; superblock after data */
1868 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1869 sb_start &= ~(sector_t)(4*2 - 1);
1870 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1871 if (!num_sectors || num_sectors > max_sectors)
1872 num_sectors = max_sectors;
1873 rdev->sb_start = sb_start;
1875 sb = page_address(rdev->sb_page);
1876 sb->data_size = cpu_to_le64(num_sectors);
1877 sb->super_offset = cpu_to_le64(rdev->sb_start);
1878 sb->sb_csum = calc_sb_1_csum(sb);
1879 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1881 md_super_wait(rdev->mddev);
1887 super_1_allow_new_offset(struct md_rdev *rdev,
1888 unsigned long long new_offset)
1890 /* All necessary checks on new >= old have been done */
1891 struct bitmap *bitmap;
1892 if (new_offset >= rdev->data_offset)
1895 /* with 1.0 metadata, there is no metadata to tread on
1896 * so we can always move back */
1897 if (rdev->mddev->minor_version == 0)
1900 /* otherwise we must be sure not to step on
1901 * any metadata, so stay:
1902 * 36K beyond start of superblock
1903 * beyond end of badblocks
1904 * beyond write-intent bitmap
1906 if (rdev->sb_start + (32+4)*2 > new_offset)
1908 bitmap = rdev->mddev->bitmap;
1909 if (bitmap && !rdev->mddev->bitmap_info.file &&
1910 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1911 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1913 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1919 static struct super_type super_types[] = {
1922 .owner = THIS_MODULE,
1923 .load_super = super_90_load,
1924 .validate_super = super_90_validate,
1925 .sync_super = super_90_sync,
1926 .rdev_size_change = super_90_rdev_size_change,
1927 .allow_new_offset = super_90_allow_new_offset,
1931 .owner = THIS_MODULE,
1932 .load_super = super_1_load,
1933 .validate_super = super_1_validate,
1934 .sync_super = super_1_sync,
1935 .rdev_size_change = super_1_rdev_size_change,
1936 .allow_new_offset = super_1_allow_new_offset,
1940 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1942 if (mddev->sync_super) {
1943 mddev->sync_super(mddev, rdev);
1947 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1949 super_types[mddev->major_version].sync_super(mddev, rdev);
1952 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1954 struct md_rdev *rdev, *rdev2;
1957 rdev_for_each_rcu(rdev, mddev1) {
1958 if (test_bit(Faulty, &rdev->flags) ||
1959 test_bit(Journal, &rdev->flags) ||
1960 rdev->raid_disk == -1)
1962 rdev_for_each_rcu(rdev2, mddev2) {
1963 if (test_bit(Faulty, &rdev2->flags) ||
1964 test_bit(Journal, &rdev2->flags) ||
1965 rdev2->raid_disk == -1)
1967 if (rdev->bdev->bd_contains ==
1968 rdev2->bdev->bd_contains) {
1978 static LIST_HEAD(pending_raid_disks);
1981 * Try to register data integrity profile for an mddev
1983 * This is called when an array is started and after a disk has been kicked
1984 * from the array. It only succeeds if all working and active component devices
1985 * are integrity capable with matching profiles.
1987 int md_integrity_register(struct mddev *mddev)
1989 struct md_rdev *rdev, *reference = NULL;
1991 if (list_empty(&mddev->disks))
1992 return 0; /* nothing to do */
1993 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1994 return 0; /* shouldn't register, or already is */
1995 rdev_for_each(rdev, mddev) {
1996 /* skip spares and non-functional disks */
1997 if (test_bit(Faulty, &rdev->flags))
1999 if (rdev->raid_disk < 0)
2002 /* Use the first rdev as the reference */
2006 /* does this rdev's profile match the reference profile? */
2007 if (blk_integrity_compare(reference->bdev->bd_disk,
2008 rdev->bdev->bd_disk) < 0)
2011 if (!reference || !bdev_get_integrity(reference->bdev))
2014 * All component devices are integrity capable and have matching
2015 * profiles, register the common profile for the md device.
2017 blk_integrity_register(mddev->gendisk,
2018 bdev_get_integrity(reference->bdev));
2020 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2021 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2022 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2028 EXPORT_SYMBOL(md_integrity_register);
2031 * Attempt to add an rdev, but only if it is consistent with the current
2034 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2036 struct blk_integrity *bi_rdev;
2037 struct blk_integrity *bi_mddev;
2038 char name[BDEVNAME_SIZE];
2040 if (!mddev->gendisk)
2043 bi_rdev = bdev_get_integrity(rdev->bdev);
2044 bi_mddev = blk_get_integrity(mddev->gendisk);
2046 if (!bi_mddev) /* nothing to do */
2049 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2050 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2051 mdname(mddev), bdevname(rdev->bdev, name));
2057 EXPORT_SYMBOL(md_integrity_add_rdev);
2059 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2061 char b[BDEVNAME_SIZE];
2065 /* prevent duplicates */
2066 if (find_rdev(mddev, rdev->bdev->bd_dev))
2069 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2070 if (!test_bit(Journal, &rdev->flags) &&
2072 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2074 /* Cannot change size, so fail
2075 * If mddev->level <= 0, then we don't care
2076 * about aligning sizes (e.g. linear)
2078 if (mddev->level > 0)
2081 mddev->dev_sectors = rdev->sectors;
2084 /* Verify rdev->desc_nr is unique.
2085 * If it is -1, assign a free number, else
2086 * check number is not in use
2089 if (rdev->desc_nr < 0) {
2092 choice = mddev->raid_disks;
2093 while (md_find_rdev_nr_rcu(mddev, choice))
2095 rdev->desc_nr = choice;
2097 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2103 if (!test_bit(Journal, &rdev->flags) &&
2104 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2105 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2106 mdname(mddev), mddev->max_disks);
2109 bdevname(rdev->bdev,b);
2110 strreplace(b, '/', '!');
2112 rdev->mddev = mddev;
2113 printk(KERN_INFO "md: bind<%s>\n", b);
2115 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2118 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2119 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2120 /* failure here is OK */;
2121 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2123 list_add_rcu(&rdev->same_set, &mddev->disks);
2124 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2126 /* May as well allow recovery to be retried once */
2127 mddev->recovery_disabled++;
2132 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2137 static void md_delayed_delete(struct work_struct *ws)
2139 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2140 kobject_del(&rdev->kobj);
2141 kobject_put(&rdev->kobj);
2144 static void unbind_rdev_from_array(struct md_rdev *rdev)
2146 char b[BDEVNAME_SIZE];
2148 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2149 list_del_rcu(&rdev->same_set);
2150 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2152 sysfs_remove_link(&rdev->kobj, "block");
2153 sysfs_put(rdev->sysfs_state);
2154 rdev->sysfs_state = NULL;
2155 rdev->badblocks.count = 0;
2156 /* We need to delay this, otherwise we can deadlock when
2157 * writing to 'remove' to "dev/state". We also need
2158 * to delay it due to rcu usage.
2161 INIT_WORK(&rdev->del_work, md_delayed_delete);
2162 kobject_get(&rdev->kobj);
2163 queue_work(md_misc_wq, &rdev->del_work);
2167 * prevent the device from being mounted, repartitioned or
2168 * otherwise reused by a RAID array (or any other kernel
2169 * subsystem), by bd_claiming the device.
2171 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2174 struct block_device *bdev;
2175 char b[BDEVNAME_SIZE];
2177 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2178 shared ? (struct md_rdev *)lock_rdev : rdev);
2180 printk(KERN_ERR "md: could not open %s.\n",
2181 __bdevname(dev, b));
2182 return PTR_ERR(bdev);
2188 static void unlock_rdev(struct md_rdev *rdev)
2190 struct block_device *bdev = rdev->bdev;
2192 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2195 void md_autodetect_dev(dev_t dev);
2197 static void export_rdev(struct md_rdev *rdev)
2199 char b[BDEVNAME_SIZE];
2201 printk(KERN_INFO "md: export_rdev(%s)\n",
2202 bdevname(rdev->bdev,b));
2203 md_rdev_clear(rdev);
2205 if (test_bit(AutoDetected, &rdev->flags))
2206 md_autodetect_dev(rdev->bdev->bd_dev);
2209 kobject_put(&rdev->kobj);
2212 void md_kick_rdev_from_array(struct md_rdev *rdev)
2214 unbind_rdev_from_array(rdev);
2217 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2219 static void export_array(struct mddev *mddev)
2221 struct md_rdev *rdev;
2223 while (!list_empty(&mddev->disks)) {
2224 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2226 md_kick_rdev_from_array(rdev);
2228 mddev->raid_disks = 0;
2229 mddev->major_version = 0;
2232 static void sync_sbs(struct mddev *mddev, int nospares)
2234 /* Update each superblock (in-memory image), but
2235 * if we are allowed to, skip spares which already
2236 * have the right event counter, or have one earlier
2237 * (which would mean they aren't being marked as dirty
2238 * with the rest of the array)
2240 struct md_rdev *rdev;
2241 rdev_for_each(rdev, mddev) {
2242 if (rdev->sb_events == mddev->events ||
2244 rdev->raid_disk < 0 &&
2245 rdev->sb_events+1 == mddev->events)) {
2246 /* Don't update this superblock */
2247 rdev->sb_loaded = 2;
2249 sync_super(mddev, rdev);
2250 rdev->sb_loaded = 1;
2255 static bool does_sb_need_changing(struct mddev *mddev)
2257 struct md_rdev *rdev;
2258 struct mdp_superblock_1 *sb;
2261 /* Find a good rdev */
2262 rdev_for_each(rdev, mddev)
2263 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2266 /* No good device found. */
2270 sb = page_address(rdev->sb_page);
2271 /* Check if a device has become faulty or a spare become active */
2272 rdev_for_each(rdev, mddev) {
2273 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2274 /* Device activated? */
2275 if (role == 0xffff && rdev->raid_disk >=0 &&
2276 !test_bit(Faulty, &rdev->flags))
2278 /* Device turned faulty? */
2279 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2283 /* Check if any mddev parameters have changed */
2284 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2285 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2286 (mddev->layout != le32_to_cpu(sb->layout)) ||
2287 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2288 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2294 void md_update_sb(struct mddev *mddev, int force_change)
2296 struct md_rdev *rdev;
2299 int any_badblocks_changed = 0;
2304 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2309 if (mddev_is_clustered(mddev)) {
2310 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2312 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2314 ret = md_cluster_ops->metadata_update_start(mddev);
2315 /* Has someone else has updated the sb */
2316 if (!does_sb_need_changing(mddev)) {
2318 md_cluster_ops->metadata_update_cancel(mddev);
2319 bit_clear_unless(&mddev->flags, BIT(MD_CHANGE_PENDING),
2320 BIT(MD_CHANGE_DEVS) |
2321 BIT(MD_CHANGE_CLEAN));
2326 /* First make sure individual recovery_offsets are correct */
2327 rdev_for_each(rdev, mddev) {
2328 if (rdev->raid_disk >= 0 &&
2329 mddev->delta_disks >= 0 &&
2330 !test_bit(Journal, &rdev->flags) &&
2331 !test_bit(In_sync, &rdev->flags) &&
2332 mddev->curr_resync_completed > rdev->recovery_offset)
2333 rdev->recovery_offset = mddev->curr_resync_completed;
2336 if (!mddev->persistent) {
2337 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2338 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2339 if (!mddev->external) {
2340 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2341 rdev_for_each(rdev, mddev) {
2342 if (rdev->badblocks.changed) {
2343 rdev->badblocks.changed = 0;
2344 ack_all_badblocks(&rdev->badblocks);
2345 md_error(mddev, rdev);
2347 clear_bit(Blocked, &rdev->flags);
2348 clear_bit(BlockedBadBlocks, &rdev->flags);
2349 wake_up(&rdev->blocked_wait);
2352 wake_up(&mddev->sb_wait);
2356 spin_lock(&mddev->lock);
2358 mddev->utime = ktime_get_real_seconds();
2360 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2362 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2363 /* just a clean<-> dirty transition, possibly leave spares alone,
2364 * though if events isn't the right even/odd, we will have to do
2370 if (mddev->degraded)
2371 /* If the array is degraded, then skipping spares is both
2372 * dangerous and fairly pointless.
2373 * Dangerous because a device that was removed from the array
2374 * might have a event_count that still looks up-to-date,
2375 * so it can be re-added without a resync.
2376 * Pointless because if there are any spares to skip,
2377 * then a recovery will happen and soon that array won't
2378 * be degraded any more and the spare can go back to sleep then.
2382 sync_req = mddev->in_sync;
2384 /* If this is just a dirty<->clean transition, and the array is clean
2385 * and 'events' is odd, we can roll back to the previous clean state */
2387 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2388 && mddev->can_decrease_events
2389 && mddev->events != 1) {
2391 mddev->can_decrease_events = 0;
2393 /* otherwise we have to go forward and ... */
2395 mddev->can_decrease_events = nospares;
2399 * This 64-bit counter should never wrap.
2400 * Either we are in around ~1 trillion A.C., assuming
2401 * 1 reboot per second, or we have a bug...
2403 WARN_ON(mddev->events == 0);
2405 rdev_for_each(rdev, mddev) {
2406 if (rdev->badblocks.changed)
2407 any_badblocks_changed++;
2408 if (test_bit(Faulty, &rdev->flags))
2409 set_bit(FaultRecorded, &rdev->flags);
2412 sync_sbs(mddev, nospares);
2413 spin_unlock(&mddev->lock);
2415 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2416 mdname(mddev), mddev->in_sync);
2418 bitmap_update_sb(mddev->bitmap);
2419 rdev_for_each(rdev, mddev) {
2420 char b[BDEVNAME_SIZE];
2422 if (rdev->sb_loaded != 1)
2423 continue; /* no noise on spare devices */
2425 if (!test_bit(Faulty, &rdev->flags)) {
2426 md_super_write(mddev,rdev,
2427 rdev->sb_start, rdev->sb_size,
2429 pr_debug("md: (write) %s's sb offset: %llu\n",
2430 bdevname(rdev->bdev, b),
2431 (unsigned long long)rdev->sb_start);
2432 rdev->sb_events = mddev->events;
2433 if (rdev->badblocks.size) {
2434 md_super_write(mddev, rdev,
2435 rdev->badblocks.sector,
2436 rdev->badblocks.size << 9,
2438 rdev->badblocks.size = 0;
2442 pr_debug("md: %s (skipping faulty)\n",
2443 bdevname(rdev->bdev, b));
2445 if (mddev->level == LEVEL_MULTIPATH)
2446 /* only need to write one superblock... */
2449 md_super_wait(mddev);
2450 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2452 if (mddev_is_clustered(mddev) && ret == 0)
2453 md_cluster_ops->metadata_update_finish(mddev);
2455 if (mddev->in_sync != sync_req ||
2456 !bit_clear_unless(&mddev->flags, BIT(MD_CHANGE_PENDING),
2457 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_CLEAN)))
2458 /* have to write it out again */
2460 wake_up(&mddev->sb_wait);
2461 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2462 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2464 rdev_for_each(rdev, mddev) {
2465 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2466 clear_bit(Blocked, &rdev->flags);
2468 if (any_badblocks_changed)
2469 ack_all_badblocks(&rdev->badblocks);
2470 clear_bit(BlockedBadBlocks, &rdev->flags);
2471 wake_up(&rdev->blocked_wait);
2474 EXPORT_SYMBOL(md_update_sb);
2476 static int add_bound_rdev(struct md_rdev *rdev)
2478 struct mddev *mddev = rdev->mddev;
2480 bool add_journal = test_bit(Journal, &rdev->flags);
2482 if (!mddev->pers->hot_remove_disk || add_journal) {
2483 /* If there is hot_add_disk but no hot_remove_disk
2484 * then added disks for geometry changes,
2485 * and should be added immediately.
2487 super_types[mddev->major_version].
2488 validate_super(mddev, rdev);
2490 mddev_suspend(mddev);
2491 err = mddev->pers->hot_add_disk(mddev, rdev);
2493 mddev_resume(mddev);
2495 md_kick_rdev_from_array(rdev);
2499 sysfs_notify_dirent_safe(rdev->sysfs_state);
2501 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2502 if (mddev->degraded)
2503 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2504 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2505 md_new_event(mddev);
2506 md_wakeup_thread(mddev->thread);
2510 /* words written to sysfs files may, or may not, be \n terminated.
2511 * We want to accept with case. For this we use cmd_match.
2513 static int cmd_match(const char *cmd, const char *str)
2515 /* See if cmd, written into a sysfs file, matches
2516 * str. They must either be the same, or cmd can
2517 * have a trailing newline
2519 while (*cmd && *str && *cmd == *str) {
2530 struct rdev_sysfs_entry {
2531 struct attribute attr;
2532 ssize_t (*show)(struct md_rdev *, char *);
2533 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2537 state_show(struct md_rdev *rdev, char *page)
2541 unsigned long flags = ACCESS_ONCE(rdev->flags);
2543 if (test_bit(Faulty, &flags) ||
2544 rdev->badblocks.unacked_exist) {
2545 len+= sprintf(page+len, "%sfaulty",sep);
2548 if (test_bit(In_sync, &flags)) {
2549 len += sprintf(page+len, "%sin_sync",sep);
2552 if (test_bit(Journal, &flags)) {
2553 len += sprintf(page+len, "%sjournal",sep);
2556 if (test_bit(WriteMostly, &flags)) {
2557 len += sprintf(page+len, "%swrite_mostly",sep);
2560 if (test_bit(Blocked, &flags) ||
2561 (rdev->badblocks.unacked_exist
2562 && !test_bit(Faulty, &flags))) {
2563 len += sprintf(page+len, "%sblocked", sep);
2566 if (!test_bit(Faulty, &flags) &&
2567 !test_bit(Journal, &flags) &&
2568 !test_bit(In_sync, &flags)) {
2569 len += sprintf(page+len, "%sspare", sep);
2572 if (test_bit(WriteErrorSeen, &flags)) {
2573 len += sprintf(page+len, "%swrite_error", sep);
2576 if (test_bit(WantReplacement, &flags)) {
2577 len += sprintf(page+len, "%swant_replacement", sep);
2580 if (test_bit(Replacement, &flags)) {
2581 len += sprintf(page+len, "%sreplacement", sep);
2585 return len+sprintf(page+len, "\n");
2589 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2592 * faulty - simulates an error
2593 * remove - disconnects the device
2594 * writemostly - sets write_mostly
2595 * -writemostly - clears write_mostly
2596 * blocked - sets the Blocked flags
2597 * -blocked - clears the Blocked and possibly simulates an error
2598 * insync - sets Insync providing device isn't active
2599 * -insync - clear Insync for a device with a slot assigned,
2600 * so that it gets rebuilt based on bitmap
2601 * write_error - sets WriteErrorSeen
2602 * -write_error - clears WriteErrorSeen
2605 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2606 md_error(rdev->mddev, rdev);
2607 if (test_bit(Faulty, &rdev->flags))
2611 } else if (cmd_match(buf, "remove")) {
2612 if (rdev->mddev->pers) {
2613 clear_bit(Blocked, &rdev->flags);
2614 remove_and_add_spares(rdev->mddev, rdev);
2616 if (rdev->raid_disk >= 0)
2619 struct mddev *mddev = rdev->mddev;
2621 if (mddev_is_clustered(mddev))
2622 err = md_cluster_ops->remove_disk(mddev, rdev);
2625 md_kick_rdev_from_array(rdev);
2627 md_update_sb(mddev, 1);
2628 md_new_event(mddev);
2631 } else if (cmd_match(buf, "writemostly")) {
2632 set_bit(WriteMostly, &rdev->flags);
2634 } else if (cmd_match(buf, "-writemostly")) {
2635 clear_bit(WriteMostly, &rdev->flags);
2637 } else if (cmd_match(buf, "blocked")) {
2638 set_bit(Blocked, &rdev->flags);
2640 } else if (cmd_match(buf, "-blocked")) {
2641 if (!test_bit(Faulty, &rdev->flags) &&
2642 rdev->badblocks.unacked_exist) {
2643 /* metadata handler doesn't understand badblocks,
2644 * so we need to fail the device
2646 md_error(rdev->mddev, rdev);
2648 clear_bit(Blocked, &rdev->flags);
2649 clear_bit(BlockedBadBlocks, &rdev->flags);
2650 wake_up(&rdev->blocked_wait);
2651 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2652 md_wakeup_thread(rdev->mddev->thread);
2655 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2656 set_bit(In_sync, &rdev->flags);
2658 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2659 !test_bit(Journal, &rdev->flags)) {
2660 if (rdev->mddev->pers == NULL) {
2661 clear_bit(In_sync, &rdev->flags);
2662 rdev->saved_raid_disk = rdev->raid_disk;
2663 rdev->raid_disk = -1;
2666 } else if (cmd_match(buf, "write_error")) {
2667 set_bit(WriteErrorSeen, &rdev->flags);
2669 } else if (cmd_match(buf, "-write_error")) {
2670 clear_bit(WriteErrorSeen, &rdev->flags);
2672 } else if (cmd_match(buf, "want_replacement")) {
2673 /* Any non-spare device that is not a replacement can
2674 * become want_replacement at any time, but we then need to
2675 * check if recovery is needed.
2677 if (rdev->raid_disk >= 0 &&
2678 !test_bit(Journal, &rdev->flags) &&
2679 !test_bit(Replacement, &rdev->flags))
2680 set_bit(WantReplacement, &rdev->flags);
2681 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2682 md_wakeup_thread(rdev->mddev->thread);
2684 } else if (cmd_match(buf, "-want_replacement")) {
2685 /* Clearing 'want_replacement' is always allowed.
2686 * Once replacements starts it is too late though.
2689 clear_bit(WantReplacement, &rdev->flags);
2690 } else if (cmd_match(buf, "replacement")) {
2691 /* Can only set a device as a replacement when array has not
2692 * yet been started. Once running, replacement is automatic
2693 * from spares, or by assigning 'slot'.
2695 if (rdev->mddev->pers)
2698 set_bit(Replacement, &rdev->flags);
2701 } else if (cmd_match(buf, "-replacement")) {
2702 /* Similarly, can only clear Replacement before start */
2703 if (rdev->mddev->pers)
2706 clear_bit(Replacement, &rdev->flags);
2709 } else if (cmd_match(buf, "re-add")) {
2710 if (!rdev->mddev->pers)
2712 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
2713 rdev->saved_raid_disk >= 0) {
2714 /* clear_bit is performed _after_ all the devices
2715 * have their local Faulty bit cleared. If any writes
2716 * happen in the meantime in the local node, they
2717 * will land in the local bitmap, which will be synced
2718 * by this node eventually
2720 if (!mddev_is_clustered(rdev->mddev) ||
2721 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2722 clear_bit(Faulty, &rdev->flags);
2723 err = add_bound_rdev(rdev);
2729 sysfs_notify_dirent_safe(rdev->sysfs_state);
2730 return err ? err : len;
2732 static struct rdev_sysfs_entry rdev_state =
2733 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2736 errors_show(struct md_rdev *rdev, char *page)
2738 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2742 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2747 rv = kstrtouint(buf, 10, &n);
2750 atomic_set(&rdev->corrected_errors, n);
2753 static struct rdev_sysfs_entry rdev_errors =
2754 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2757 slot_show(struct md_rdev *rdev, char *page)
2759 if (test_bit(Journal, &rdev->flags))
2760 return sprintf(page, "journal\n");
2761 else if (rdev->raid_disk < 0)
2762 return sprintf(page, "none\n");
2764 return sprintf(page, "%d\n", rdev->raid_disk);
2768 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2773 if (test_bit(Journal, &rdev->flags))
2775 if (strncmp(buf, "none", 4)==0)
2778 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2782 if (rdev->mddev->pers && slot == -1) {
2783 /* Setting 'slot' on an active array requires also
2784 * updating the 'rd%d' link, and communicating
2785 * with the personality with ->hot_*_disk.
2786 * For now we only support removing
2787 * failed/spare devices. This normally happens automatically,
2788 * but not when the metadata is externally managed.
2790 if (rdev->raid_disk == -1)
2792 /* personality does all needed checks */
2793 if (rdev->mddev->pers->hot_remove_disk == NULL)
2795 clear_bit(Blocked, &rdev->flags);
2796 remove_and_add_spares(rdev->mddev, rdev);
2797 if (rdev->raid_disk >= 0)
2799 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2800 md_wakeup_thread(rdev->mddev->thread);
2801 } else if (rdev->mddev->pers) {
2802 /* Activating a spare .. or possibly reactivating
2803 * if we ever get bitmaps working here.
2807 if (rdev->raid_disk != -1)
2810 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2813 if (rdev->mddev->pers->hot_add_disk == NULL)
2816 if (slot >= rdev->mddev->raid_disks &&
2817 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2820 rdev->raid_disk = slot;
2821 if (test_bit(In_sync, &rdev->flags))
2822 rdev->saved_raid_disk = slot;
2824 rdev->saved_raid_disk = -1;
2825 clear_bit(In_sync, &rdev->flags);
2826 clear_bit(Bitmap_sync, &rdev->flags);
2827 err = rdev->mddev->pers->
2828 hot_add_disk(rdev->mddev, rdev);
2830 rdev->raid_disk = -1;
2833 sysfs_notify_dirent_safe(rdev->sysfs_state);
2834 if (sysfs_link_rdev(rdev->mddev, rdev))
2835 /* failure here is OK */;
2836 /* don't wakeup anyone, leave that to userspace. */
2838 if (slot >= rdev->mddev->raid_disks &&
2839 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2841 rdev->raid_disk = slot;
2842 /* assume it is working */
2843 clear_bit(Faulty, &rdev->flags);
2844 clear_bit(WriteMostly, &rdev->flags);
2845 set_bit(In_sync, &rdev->flags);
2846 sysfs_notify_dirent_safe(rdev->sysfs_state);
2851 static struct rdev_sysfs_entry rdev_slot =
2852 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2855 offset_show(struct md_rdev *rdev, char *page)
2857 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2861 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2863 unsigned long long offset;
2864 if (kstrtoull(buf, 10, &offset) < 0)
2866 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2868 if (rdev->sectors && rdev->mddev->external)
2869 /* Must set offset before size, so overlap checks
2872 rdev->data_offset = offset;
2873 rdev->new_data_offset = offset;
2877 static struct rdev_sysfs_entry rdev_offset =
2878 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2880 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2882 return sprintf(page, "%llu\n",
2883 (unsigned long long)rdev->new_data_offset);
2886 static ssize_t new_offset_store(struct md_rdev *rdev,
2887 const char *buf, size_t len)
2889 unsigned long long new_offset;
2890 struct mddev *mddev = rdev->mddev;
2892 if (kstrtoull(buf, 10, &new_offset) < 0)
2895 if (mddev->sync_thread ||
2896 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2898 if (new_offset == rdev->data_offset)
2899 /* reset is always permitted */
2901 else if (new_offset > rdev->data_offset) {
2902 /* must not push array size beyond rdev_sectors */
2903 if (new_offset - rdev->data_offset
2904 + mddev->dev_sectors > rdev->sectors)
2907 /* Metadata worries about other space details. */
2909 /* decreasing the offset is inconsistent with a backwards
2912 if (new_offset < rdev->data_offset &&
2913 mddev->reshape_backwards)
2915 /* Increasing offset is inconsistent with forwards
2916 * reshape. reshape_direction should be set to
2917 * 'backwards' first.
2919 if (new_offset > rdev->data_offset &&
2920 !mddev->reshape_backwards)
2923 if (mddev->pers && mddev->persistent &&
2924 !super_types[mddev->major_version]
2925 .allow_new_offset(rdev, new_offset))
2927 rdev->new_data_offset = new_offset;
2928 if (new_offset > rdev->data_offset)
2929 mddev->reshape_backwards = 1;
2930 else if (new_offset < rdev->data_offset)
2931 mddev->reshape_backwards = 0;
2935 static struct rdev_sysfs_entry rdev_new_offset =
2936 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2939 rdev_size_show(struct md_rdev *rdev, char *page)
2941 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2944 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2946 /* check if two start/length pairs overlap */
2954 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2956 unsigned long long blocks;
2959 if (kstrtoull(buf, 10, &blocks) < 0)
2962 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2963 return -EINVAL; /* sector conversion overflow */
2966 if (new != blocks * 2)
2967 return -EINVAL; /* unsigned long long to sector_t overflow */
2974 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2976 struct mddev *my_mddev = rdev->mddev;
2977 sector_t oldsectors = rdev->sectors;
2980 if (test_bit(Journal, &rdev->flags))
2982 if (strict_blocks_to_sectors(buf, §ors) < 0)
2984 if (rdev->data_offset != rdev->new_data_offset)
2985 return -EINVAL; /* too confusing */
2986 if (my_mddev->pers && rdev->raid_disk >= 0) {
2987 if (my_mddev->persistent) {
2988 sectors = super_types[my_mddev->major_version].
2989 rdev_size_change(rdev, sectors);
2992 } else if (!sectors)
2993 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2995 if (!my_mddev->pers->resize)
2996 /* Cannot change size for RAID0 or Linear etc */
2999 if (sectors < my_mddev->dev_sectors)
3000 return -EINVAL; /* component must fit device */
3002 rdev->sectors = sectors;
3003 if (sectors > oldsectors && my_mddev->external) {
3004 /* Need to check that all other rdevs with the same
3005 * ->bdev do not overlap. 'rcu' is sufficient to walk
3006 * the rdev lists safely.
3007 * This check does not provide a hard guarantee, it
3008 * just helps avoid dangerous mistakes.
3010 struct mddev *mddev;
3012 struct list_head *tmp;
3015 for_each_mddev(mddev, tmp) {
3016 struct md_rdev *rdev2;
3018 rdev_for_each(rdev2, mddev)
3019 if (rdev->bdev == rdev2->bdev &&
3021 overlaps(rdev->data_offset, rdev->sectors,
3034 /* Someone else could have slipped in a size
3035 * change here, but doing so is just silly.
3036 * We put oldsectors back because we *know* it is
3037 * safe, and trust userspace not to race with
3040 rdev->sectors = oldsectors;
3047 static struct rdev_sysfs_entry rdev_size =
3048 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3050 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3052 unsigned long long recovery_start = rdev->recovery_offset;
3054 if (test_bit(In_sync, &rdev->flags) ||
3055 recovery_start == MaxSector)
3056 return sprintf(page, "none\n");
3058 return sprintf(page, "%llu\n", recovery_start);
3061 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3063 unsigned long long recovery_start;
3065 if (cmd_match(buf, "none"))
3066 recovery_start = MaxSector;
3067 else if (kstrtoull(buf, 10, &recovery_start))
3070 if (rdev->mddev->pers &&
3071 rdev->raid_disk >= 0)
3074 rdev->recovery_offset = recovery_start;
3075 if (recovery_start == MaxSector)
3076 set_bit(In_sync, &rdev->flags);
3078 clear_bit(In_sync, &rdev->flags);
3082 static struct rdev_sysfs_entry rdev_recovery_start =
3083 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3085 /* sysfs access to bad-blocks list.
3086 * We present two files.
3087 * 'bad-blocks' lists sector numbers and lengths of ranges that
3088 * are recorded as bad. The list is truncated to fit within
3089 * the one-page limit of sysfs.
3090 * Writing "sector length" to this file adds an acknowledged
3092 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3093 * been acknowledged. Writing to this file adds bad blocks
3094 * without acknowledging them. This is largely for testing.
3096 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3098 return badblocks_show(&rdev->badblocks, page, 0);
3100 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3102 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3103 /* Maybe that ack was all we needed */
3104 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3105 wake_up(&rdev->blocked_wait);
3108 static struct rdev_sysfs_entry rdev_bad_blocks =
3109 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3111 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3113 return badblocks_show(&rdev->badblocks, page, 1);
3115 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3117 return badblocks_store(&rdev->badblocks, page, len, 1);
3119 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3120 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3122 static struct attribute *rdev_default_attrs[] = {
3127 &rdev_new_offset.attr,
3129 &rdev_recovery_start.attr,
3130 &rdev_bad_blocks.attr,
3131 &rdev_unack_bad_blocks.attr,
3135 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3137 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3138 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3144 return entry->show(rdev, page);
3148 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3149 const char *page, size_t length)
3151 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3152 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3154 struct mddev *mddev = rdev->mddev;
3158 if (!capable(CAP_SYS_ADMIN))
3160 rv = mddev ? mddev_lock(mddev): -EBUSY;
3162 if (rdev->mddev == NULL)
3165 rv = entry->store(rdev, page, length);
3166 mddev_unlock(mddev);
3171 static void rdev_free(struct kobject *ko)
3173 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3176 static const struct sysfs_ops rdev_sysfs_ops = {
3177 .show = rdev_attr_show,
3178 .store = rdev_attr_store,
3180 static struct kobj_type rdev_ktype = {
3181 .release = rdev_free,
3182 .sysfs_ops = &rdev_sysfs_ops,
3183 .default_attrs = rdev_default_attrs,
3186 int md_rdev_init(struct md_rdev *rdev)
3189 rdev->saved_raid_disk = -1;
3190 rdev->raid_disk = -1;
3192 rdev->data_offset = 0;
3193 rdev->new_data_offset = 0;
3194 rdev->sb_events = 0;
3195 rdev->last_read_error = 0;
3196 rdev->sb_loaded = 0;
3197 rdev->bb_page = NULL;
3198 atomic_set(&rdev->nr_pending, 0);
3199 atomic_set(&rdev->read_errors, 0);
3200 atomic_set(&rdev->corrected_errors, 0);
3202 INIT_LIST_HEAD(&rdev->same_set);
3203 init_waitqueue_head(&rdev->blocked_wait);
3205 /* Add space to store bad block list.
3206 * This reserves the space even on arrays where it cannot
3207 * be used - I wonder if that matters
3209 return badblocks_init(&rdev->badblocks, 0);
3211 EXPORT_SYMBOL_GPL(md_rdev_init);
3213 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3215 * mark the device faulty if:
3217 * - the device is nonexistent (zero size)
3218 * - the device has no valid superblock
3220 * a faulty rdev _never_ has rdev->sb set.
3222 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3224 char b[BDEVNAME_SIZE];
3226 struct md_rdev *rdev;
3229 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3231 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3232 return ERR_PTR(-ENOMEM);
3235 err = md_rdev_init(rdev);
3238 err = alloc_disk_sb(rdev);
3242 err = lock_rdev(rdev, newdev, super_format == -2);
3246 kobject_init(&rdev->kobj, &rdev_ktype);
3248 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3251 "md: %s has zero or unknown size, marking faulty!\n",
3252 bdevname(rdev->bdev,b));
3257 if (super_format >= 0) {
3258 err = super_types[super_format].
3259 load_super(rdev, NULL, super_minor);
3260 if (err == -EINVAL) {
3262 "md: %s does not have a valid v%d.%d "
3263 "superblock, not importing!\n",
3264 bdevname(rdev->bdev,b),
3265 super_format, super_minor);
3270 "md: could not read %s's sb, not importing!\n",
3271 bdevname(rdev->bdev,b));
3281 md_rdev_clear(rdev);
3283 return ERR_PTR(err);
3287 * Check a full RAID array for plausibility
3290 static void analyze_sbs(struct mddev *mddev)
3293 struct md_rdev *rdev, *freshest, *tmp;
3294 char b[BDEVNAME_SIZE];
3297 rdev_for_each_safe(rdev, tmp, mddev)
3298 switch (super_types[mddev->major_version].
3299 load_super(rdev, freshest, mddev->minor_version)) {
3307 "md: fatal superblock inconsistency in %s"
3308 " -- removing from array\n",
3309 bdevname(rdev->bdev,b));
3310 md_kick_rdev_from_array(rdev);
3313 super_types[mddev->major_version].
3314 validate_super(mddev, freshest);
3317 rdev_for_each_safe(rdev, tmp, mddev) {
3318 if (mddev->max_disks &&
3319 (rdev->desc_nr >= mddev->max_disks ||
3320 i > mddev->max_disks)) {
3322 "md: %s: %s: only %d devices permitted\n",
3323 mdname(mddev), bdevname(rdev->bdev, b),
3325 md_kick_rdev_from_array(rdev);
3328 if (rdev != freshest) {
3329 if (super_types[mddev->major_version].
3330 validate_super(mddev, rdev)) {
3331 printk(KERN_WARNING "md: kicking non-fresh %s"
3333 bdevname(rdev->bdev,b));
3334 md_kick_rdev_from_array(rdev);
3338 if (mddev->level == LEVEL_MULTIPATH) {
3339 rdev->desc_nr = i++;
3340 rdev->raid_disk = rdev->desc_nr;
3341 set_bit(In_sync, &rdev->flags);
3342 } else if (rdev->raid_disk >=
3343 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3344 !test_bit(Journal, &rdev->flags)) {
3345 rdev->raid_disk = -1;
3346 clear_bit(In_sync, &rdev->flags);
3351 /* Read a fixed-point number.
3352 * Numbers in sysfs attributes should be in "standard" units where
3353 * possible, so time should be in seconds.
3354 * However we internally use a a much smaller unit such as
3355 * milliseconds or jiffies.
3356 * This function takes a decimal number with a possible fractional
3357 * component, and produces an integer which is the result of
3358 * multiplying that number by 10^'scale'.
3359 * all without any floating-point arithmetic.
3361 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3363 unsigned long result = 0;
3365 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3368 else if (decimals < scale) {
3371 result = result * 10 + value;
3383 while (decimals < scale) {
3392 safe_delay_show(struct mddev *mddev, char *page)
3394 int msec = (mddev->safemode_delay*1000)/HZ;
3395 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3398 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3402 if (mddev_is_clustered(mddev)) {
3403 pr_info("md: Safemode is disabled for clustered mode\n");
3407 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3410 mddev->safemode_delay = 0;
3412 unsigned long old_delay = mddev->safemode_delay;
3413 unsigned long new_delay = (msec*HZ)/1000;
3417 mddev->safemode_delay = new_delay;
3418 if (new_delay < old_delay || old_delay == 0)
3419 mod_timer(&mddev->safemode_timer, jiffies+1);
3423 static struct md_sysfs_entry md_safe_delay =
3424 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3427 level_show(struct mddev *mddev, char *page)
3429 struct md_personality *p;
3431 spin_lock(&mddev->lock);
3434 ret = sprintf(page, "%s\n", p->name);
3435 else if (mddev->clevel[0])
3436 ret = sprintf(page, "%s\n", mddev->clevel);
3437 else if (mddev->level != LEVEL_NONE)
3438 ret = sprintf(page, "%d\n", mddev->level);
3441 spin_unlock(&mddev->lock);
3446 level_store(struct mddev *mddev, const char *buf, size_t len)
3451 struct md_personality *pers, *oldpers;
3453 void *priv, *oldpriv;
3454 struct md_rdev *rdev;
3456 if (slen == 0 || slen >= sizeof(clevel))
3459 rv = mddev_lock(mddev);
3463 if (mddev->pers == NULL) {
3464 strncpy(mddev->clevel, buf, slen);
3465 if (mddev->clevel[slen-1] == '\n')
3467 mddev->clevel[slen] = 0;
3468 mddev->level = LEVEL_NONE;
3476 /* request to change the personality. Need to ensure:
3477 * - array is not engaged in resync/recovery/reshape
3478 * - old personality can be suspended
3479 * - new personality will access other array.
3483 if (mddev->sync_thread ||
3484 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3485 mddev->reshape_position != MaxSector ||
3486 mddev->sysfs_active)
3490 if (!mddev->pers->quiesce) {
3491 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3492 mdname(mddev), mddev->pers->name);
3496 /* Now find the new personality */
3497 strncpy(clevel, buf, slen);
3498 if (clevel[slen-1] == '\n')
3501 if (kstrtol(clevel, 10, &level))
3504 if (request_module("md-%s", clevel) != 0)
3505 request_module("md-level-%s", clevel);
3506 spin_lock(&pers_lock);
3507 pers = find_pers(level, clevel);
3508 if (!pers || !try_module_get(pers->owner)) {
3509 spin_unlock(&pers_lock);
3510 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3514 spin_unlock(&pers_lock);
3516 if (pers == mddev->pers) {
3517 /* Nothing to do! */
3518 module_put(pers->owner);
3522 if (!pers->takeover) {
3523 module_put(pers->owner);
3524 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3525 mdname(mddev), clevel);
3530 rdev_for_each(rdev, mddev)
3531 rdev->new_raid_disk = rdev->raid_disk;
3533 /* ->takeover must set new_* and/or delta_disks
3534 * if it succeeds, and may set them when it fails.
3536 priv = pers->takeover(mddev);
3538 mddev->new_level = mddev->level;
3539 mddev->new_layout = mddev->layout;
3540 mddev->new_chunk_sectors = mddev->chunk_sectors;
3541 mddev->raid_disks -= mddev->delta_disks;
3542 mddev->delta_disks = 0;
3543 mddev->reshape_backwards = 0;
3544 module_put(pers->owner);
3545 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3546 mdname(mddev), clevel);
3551 /* Looks like we have a winner */
3552 mddev_suspend(mddev);
3553 mddev_detach(mddev);
3555 spin_lock(&mddev->lock);
3556 oldpers = mddev->pers;
3557 oldpriv = mddev->private;
3559 mddev->private = priv;
3560 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3561 mddev->level = mddev->new_level;
3562 mddev->layout = mddev->new_layout;
3563 mddev->chunk_sectors = mddev->new_chunk_sectors;
3564 mddev->delta_disks = 0;
3565 mddev->reshape_backwards = 0;
3566 mddev->degraded = 0;
3567 spin_unlock(&mddev->lock);
3569 if (oldpers->sync_request == NULL &&
3571 /* We are converting from a no-redundancy array
3572 * to a redundancy array and metadata is managed
3573 * externally so we need to be sure that writes
3574 * won't block due to a need to transition
3576 * until external management is started.
3579 mddev->safemode_delay = 0;
3580 mddev->safemode = 0;
3583 oldpers->free(mddev, oldpriv);
3585 if (oldpers->sync_request == NULL &&
3586 pers->sync_request != NULL) {
3587 /* need to add the md_redundancy_group */
3588 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3590 "md: cannot register extra attributes for %s\n",
3592 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3594 if (oldpers->sync_request != NULL &&
3595 pers->sync_request == NULL) {
3596 /* need to remove the md_redundancy_group */
3597 if (mddev->to_remove == NULL)
3598 mddev->to_remove = &md_redundancy_group;
3601 module_put(oldpers->owner);
3603 rdev_for_each(rdev, mddev) {
3604 if (rdev->raid_disk < 0)
3606 if (rdev->new_raid_disk >= mddev->raid_disks)
3607 rdev->new_raid_disk = -1;
3608 if (rdev->new_raid_disk == rdev->raid_disk)
3610 sysfs_unlink_rdev(mddev, rdev);
3612 rdev_for_each(rdev, mddev) {
3613 if (rdev->raid_disk < 0)
3615 if (rdev->new_raid_disk == rdev->raid_disk)
3617 rdev->raid_disk = rdev->new_raid_disk;
3618 if (rdev->raid_disk < 0)
3619 clear_bit(In_sync, &rdev->flags);
3621 if (sysfs_link_rdev(mddev, rdev))
3622 printk(KERN_WARNING "md: cannot register rd%d"
3623 " for %s after level change\n",
3624 rdev->raid_disk, mdname(mddev));
3628 if (pers->sync_request == NULL) {
3629 /* this is now an array without redundancy, so
3630 * it must always be in_sync
3633 del_timer_sync(&mddev->safemode_timer);
3635 blk_set_stacking_limits(&mddev->queue->limits);
3637 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3638 mddev_resume(mddev);
3640 md_update_sb(mddev, 1);
3641 sysfs_notify(&mddev->kobj, NULL, "level");
3642 md_new_event(mddev);
3645 mddev_unlock(mddev);
3649 static struct md_sysfs_entry md_level =
3650 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3653 layout_show(struct mddev *mddev, char *page)
3655 /* just a number, not meaningful for all levels */
3656 if (mddev->reshape_position != MaxSector &&
3657 mddev->layout != mddev->new_layout)
3658 return sprintf(page, "%d (%d)\n",
3659 mddev->new_layout, mddev->layout);
3660 return sprintf(page, "%d\n", mddev->layout);
3664 layout_store(struct mddev *mddev, const char *buf, size_t len)
3669 err = kstrtouint(buf, 10, &n);
3672 err = mddev_lock(mddev);
3677 if (mddev->pers->check_reshape == NULL)
3682 mddev->new_layout = n;
3683 err = mddev->pers->check_reshape(mddev);
3685 mddev->new_layout = mddev->layout;
3688 mddev->new_layout = n;
3689 if (mddev->reshape_position == MaxSector)
3692 mddev_unlock(mddev);
3695 static struct md_sysfs_entry md_layout =
3696 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3699 raid_disks_show(struct mddev *mddev, char *page)
3701 if (mddev->raid_disks == 0)
3703 if (mddev->reshape_position != MaxSector &&
3704 mddev->delta_disks != 0)
3705 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3706 mddev->raid_disks - mddev->delta_disks);
3707 return sprintf(page, "%d\n", mddev->raid_disks);
3710 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3713 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3718 err = kstrtouint(buf, 10, &n);
3722 err = mddev_lock(mddev);
3726 err = update_raid_disks(mddev, n);
3727 else if (mddev->reshape_position != MaxSector) {
3728 struct md_rdev *rdev;
3729 int olddisks = mddev->raid_disks - mddev->delta_disks;
3732 rdev_for_each(rdev, mddev) {
3734 rdev->data_offset < rdev->new_data_offset)
3737 rdev->data_offset > rdev->new_data_offset)
3741 mddev->delta_disks = n - olddisks;
3742 mddev->raid_disks = n;
3743 mddev->reshape_backwards = (mddev->delta_disks < 0);
3745 mddev->raid_disks = n;
3747 mddev_unlock(mddev);
3748 return err ? err : len;
3750 static struct md_sysfs_entry md_raid_disks =
3751 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3754 chunk_size_show(struct mddev *mddev, char *page)
3756 if (mddev->reshape_position != MaxSector &&
3757 mddev->chunk_sectors != mddev->new_chunk_sectors)
3758 return sprintf(page, "%d (%d)\n",
3759 mddev->new_chunk_sectors << 9,
3760 mddev->chunk_sectors << 9);
3761 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3765 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3770 err = kstrtoul(buf, 10, &n);
3774 err = mddev_lock(mddev);
3778 if (mddev->pers->check_reshape == NULL)
3783 mddev->new_chunk_sectors = n >> 9;
3784 err = mddev->pers->check_reshape(mddev);
3786 mddev->new_chunk_sectors = mddev->chunk_sectors;
3789 mddev->new_chunk_sectors = n >> 9;
3790 if (mddev->reshape_position == MaxSector)
3791 mddev->chunk_sectors = n >> 9;
3793 mddev_unlock(mddev);
3796 static struct md_sysfs_entry md_chunk_size =
3797 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3800 resync_start_show(struct mddev *mddev, char *page)
3802 if (mddev->recovery_cp == MaxSector)
3803 return sprintf(page, "none\n");
3804 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3808 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3810 unsigned long long n;
3813 if (cmd_match(buf, "none"))
3816 err = kstrtoull(buf, 10, &n);
3819 if (n != (sector_t)n)
3823 err = mddev_lock(mddev);
3826 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3830 mddev->recovery_cp = n;
3832 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3834 mddev_unlock(mddev);
3837 static struct md_sysfs_entry md_resync_start =
3838 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3839 resync_start_show, resync_start_store);
3842 * The array state can be:
3845 * No devices, no size, no level
3846 * Equivalent to STOP_ARRAY ioctl
3848 * May have some settings, but array is not active
3849 * all IO results in error
3850 * When written, doesn't tear down array, but just stops it
3851 * suspended (not supported yet)
3852 * All IO requests will block. The array can be reconfigured.
3853 * Writing this, if accepted, will block until array is quiescent
3855 * no resync can happen. no superblocks get written.
3856 * write requests fail
3858 * like readonly, but behaves like 'clean' on a write request.
3860 * clean - no pending writes, but otherwise active.
3861 * When written to inactive array, starts without resync
3862 * If a write request arrives then
3863 * if metadata is known, mark 'dirty' and switch to 'active'.
3864 * if not known, block and switch to write-pending
3865 * If written to an active array that has pending writes, then fails.
3867 * fully active: IO and resync can be happening.
3868 * When written to inactive array, starts with resync
3871 * clean, but writes are blocked waiting for 'active' to be written.
3874 * like active, but no writes have been seen for a while (100msec).
3877 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3878 write_pending, active_idle, bad_word};
3879 static char *array_states[] = {
3880 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3881 "write-pending", "active-idle", NULL };
3883 static int match_word(const char *word, char **list)
3886 for (n=0; list[n]; n++)
3887 if (cmd_match(word, list[n]))
3893 array_state_show(struct mddev *mddev, char *page)
3895 enum array_state st = inactive;
3906 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3908 else if (mddev->in_sync)
3910 else if (mddev->safemode)
3916 if (list_empty(&mddev->disks) &&
3917 mddev->raid_disks == 0 &&
3918 mddev->dev_sectors == 0)
3923 return sprintf(page, "%s\n", array_states[st]);
3926 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3927 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3928 static int do_md_run(struct mddev *mddev);
3929 static int restart_array(struct mddev *mddev);
3932 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3935 enum array_state st = match_word(buf, array_states);
3937 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3938 /* don't take reconfig_mutex when toggling between
3941 spin_lock(&mddev->lock);
3943 restart_array(mddev);
3944 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3945 wake_up(&mddev->sb_wait);
3947 } else /* st == clean */ {
3948 restart_array(mddev);
3949 if (atomic_read(&mddev->writes_pending) == 0) {
3950 if (mddev->in_sync == 0) {
3952 if (mddev->safemode == 1)
3953 mddev->safemode = 0;
3954 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3961 sysfs_notify_dirent_safe(mddev->sysfs_state);
3962 spin_unlock(&mddev->lock);
3965 err = mddev_lock(mddev);
3973 /* stopping an active array */
3974 err = do_md_stop(mddev, 0, NULL);
3977 /* stopping an active array */
3979 err = do_md_stop(mddev, 2, NULL);
3981 err = 0; /* already inactive */
3984 break; /* not supported yet */
3987 err = md_set_readonly(mddev, NULL);
3990 set_disk_ro(mddev->gendisk, 1);
3991 err = do_md_run(mddev);
3997 err = md_set_readonly(mddev, NULL);
3998 else if (mddev->ro == 1)
3999 err = restart_array(mddev);
4002 set_disk_ro(mddev->gendisk, 0);
4006 err = do_md_run(mddev);
4011 err = restart_array(mddev);
4014 spin_lock(&mddev->lock);
4015 if (atomic_read(&mddev->writes_pending) == 0) {
4016 if (mddev->in_sync == 0) {
4018 if (mddev->safemode == 1)
4019 mddev->safemode = 0;
4020 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
4025 spin_unlock(&mddev->lock);
4031 err = restart_array(mddev);
4034 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4035 wake_up(&mddev->sb_wait);
4039 set_disk_ro(mddev->gendisk, 0);
4040 err = do_md_run(mddev);
4045 /* these cannot be set */
4050 if (mddev->hold_active == UNTIL_IOCTL)
4051 mddev->hold_active = 0;
4052 sysfs_notify_dirent_safe(mddev->sysfs_state);
4054 mddev_unlock(mddev);
4057 static struct md_sysfs_entry md_array_state =
4058 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4061 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4062 return sprintf(page, "%d\n",
4063 atomic_read(&mddev->max_corr_read_errors));
4067 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4072 rv = kstrtouint(buf, 10, &n);
4075 atomic_set(&mddev->max_corr_read_errors, n);
4079 static struct md_sysfs_entry max_corr_read_errors =
4080 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4081 max_corrected_read_errors_store);
4084 null_show(struct mddev *mddev, char *page)
4090 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4092 /* buf must be %d:%d\n? giving major and minor numbers */
4093 /* The new device is added to the array.
4094 * If the array has a persistent superblock, we read the
4095 * superblock to initialise info and check validity.
4096 * Otherwise, only checking done is that in bind_rdev_to_array,
4097 * which mainly checks size.
4100 int major = simple_strtoul(buf, &e, 10);
4103 struct md_rdev *rdev;
4106 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4108 minor = simple_strtoul(e+1, &e, 10);
4109 if (*e && *e != '\n')
4111 dev = MKDEV(major, minor);
4112 if (major != MAJOR(dev) ||
4113 minor != MINOR(dev))
4116 flush_workqueue(md_misc_wq);
4118 err = mddev_lock(mddev);
4121 if (mddev->persistent) {
4122 rdev = md_import_device(dev, mddev->major_version,
4123 mddev->minor_version);
4124 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4125 struct md_rdev *rdev0
4126 = list_entry(mddev->disks.next,
4127 struct md_rdev, same_set);
4128 err = super_types[mddev->major_version]
4129 .load_super(rdev, rdev0, mddev->minor_version);
4133 } else if (mddev->external)
4134 rdev = md_import_device(dev, -2, -1);
4136 rdev = md_import_device(dev, -1, -1);
4139 mddev_unlock(mddev);
4140 return PTR_ERR(rdev);
4142 err = bind_rdev_to_array(rdev, mddev);
4146 mddev_unlock(mddev);
4147 return err ? err : len;
4150 static struct md_sysfs_entry md_new_device =
4151 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4154 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4157 unsigned long chunk, end_chunk;
4160 err = mddev_lock(mddev);
4165 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4167 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4168 if (buf == end) break;
4169 if (*end == '-') { /* range */
4171 end_chunk = simple_strtoul(buf, &end, 0);
4172 if (buf == end) break;
4174 if (*end && !isspace(*end)) break;
4175 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4176 buf = skip_spaces(end);
4178 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4180 mddev_unlock(mddev);
4184 static struct md_sysfs_entry md_bitmap =
4185 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4188 size_show(struct mddev *mddev, char *page)
4190 return sprintf(page, "%llu\n",
4191 (unsigned long long)mddev->dev_sectors / 2);
4194 static int update_size(struct mddev *mddev, sector_t num_sectors);
4197 size_store(struct mddev *mddev, const char *buf, size_t len)
4199 /* If array is inactive, we can reduce the component size, but
4200 * not increase it (except from 0).
4201 * If array is active, we can try an on-line resize
4204 int err = strict_blocks_to_sectors(buf, §ors);
4208 err = mddev_lock(mddev);
4212 err = update_size(mddev, sectors);
4214 md_update_sb(mddev, 1);
4216 if (mddev->dev_sectors == 0 ||
4217 mddev->dev_sectors > sectors)
4218 mddev->dev_sectors = sectors;
4222 mddev_unlock(mddev);
4223 return err ? err : len;
4226 static struct md_sysfs_entry md_size =
4227 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4229 /* Metadata version.
4231 * 'none' for arrays with no metadata (good luck...)
4232 * 'external' for arrays with externally managed metadata,
4233 * or N.M for internally known formats
4236 metadata_show(struct mddev *mddev, char *page)
4238 if (mddev->persistent)
4239 return sprintf(page, "%d.%d\n",
4240 mddev->major_version, mddev->minor_version);
4241 else if (mddev->external)
4242 return sprintf(page, "external:%s\n", mddev->metadata_type);
4244 return sprintf(page, "none\n");
4248 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4253 /* Changing the details of 'external' metadata is
4254 * always permitted. Otherwise there must be
4255 * no devices attached to the array.
4258 err = mddev_lock(mddev);
4262 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4264 else if (!list_empty(&mddev->disks))
4268 if (cmd_match(buf, "none")) {
4269 mddev->persistent = 0;
4270 mddev->external = 0;
4271 mddev->major_version = 0;
4272 mddev->minor_version = 90;
4275 if (strncmp(buf, "external:", 9) == 0) {
4276 size_t namelen = len-9;
4277 if (namelen >= sizeof(mddev->metadata_type))
4278 namelen = sizeof(mddev->metadata_type)-1;
4279 strncpy(mddev->metadata_type, buf+9, namelen);
4280 mddev->metadata_type[namelen] = 0;
4281 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4282 mddev->metadata_type[--namelen] = 0;
4283 mddev->persistent = 0;
4284 mddev->external = 1;
4285 mddev->major_version = 0;
4286 mddev->minor_version = 90;
4289 major = simple_strtoul(buf, &e, 10);
4291 if (e==buf || *e != '.')
4294 minor = simple_strtoul(buf, &e, 10);
4295 if (e==buf || (*e && *e != '\n') )
4298 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4300 mddev->major_version = major;
4301 mddev->minor_version = minor;
4302 mddev->persistent = 1;
4303 mddev->external = 0;
4306 mddev_unlock(mddev);
4310 static struct md_sysfs_entry md_metadata =
4311 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4314 action_show(struct mddev *mddev, char *page)
4316 char *type = "idle";
4317 unsigned long recovery = mddev->recovery;
4318 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4320 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4321 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4322 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4324 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4325 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4327 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4331 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4333 else if (mddev->reshape_position != MaxSector)
4336 return sprintf(page, "%s\n", type);
4340 action_store(struct mddev *mddev, const char *page, size_t len)
4342 if (!mddev->pers || !mddev->pers->sync_request)
4346 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4347 if (cmd_match(page, "frozen"))
4348 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4350 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4351 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4352 mddev_lock(mddev) == 0) {
4353 flush_workqueue(md_misc_wq);
4354 if (mddev->sync_thread) {
4355 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4356 md_reap_sync_thread(mddev);
4358 mddev_unlock(mddev);
4360 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4362 else if (cmd_match(page, "resync"))
4363 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4364 else if (cmd_match(page, "recover")) {
4365 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4366 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4367 } else if (cmd_match(page, "reshape")) {
4369 if (mddev->pers->start_reshape == NULL)
4371 err = mddev_lock(mddev);
4373 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4376 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4377 err = mddev->pers->start_reshape(mddev);
4379 mddev_unlock(mddev);
4383 sysfs_notify(&mddev->kobj, NULL, "degraded");
4385 if (cmd_match(page, "check"))
4386 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4387 else if (!cmd_match(page, "repair"))
4389 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4390 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4391 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4393 if (mddev->ro == 2) {
4394 /* A write to sync_action is enough to justify
4395 * canceling read-auto mode
4398 md_wakeup_thread(mddev->sync_thread);
4400 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4401 md_wakeup_thread(mddev->thread);
4402 sysfs_notify_dirent_safe(mddev->sysfs_action);
4406 static struct md_sysfs_entry md_scan_mode =
4407 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4410 last_sync_action_show(struct mddev *mddev, char *page)
4412 return sprintf(page, "%s\n", mddev->last_sync_action);
4415 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4418 mismatch_cnt_show(struct mddev *mddev, char *page)
4420 return sprintf(page, "%llu\n",
4421 (unsigned long long)
4422 atomic64_read(&mddev->resync_mismatches));
4425 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4428 sync_min_show(struct mddev *mddev, char *page)
4430 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4431 mddev->sync_speed_min ? "local": "system");
4435 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4440 if (strncmp(buf, "system", 6)==0) {
4443 rv = kstrtouint(buf, 10, &min);
4449 mddev->sync_speed_min = min;
4453 static struct md_sysfs_entry md_sync_min =
4454 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4457 sync_max_show(struct mddev *mddev, char *page)
4459 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4460 mddev->sync_speed_max ? "local": "system");
4464 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4469 if (strncmp(buf, "system", 6)==0) {
4472 rv = kstrtouint(buf, 10, &max);
4478 mddev->sync_speed_max = max;
4482 static struct md_sysfs_entry md_sync_max =
4483 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4486 degraded_show(struct mddev *mddev, char *page)
4488 return sprintf(page, "%d\n", mddev->degraded);
4490 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4493 sync_force_parallel_show(struct mddev *mddev, char *page)
4495 return sprintf(page, "%d\n", mddev->parallel_resync);
4499 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4503 if (kstrtol(buf, 10, &n))
4506 if (n != 0 && n != 1)
4509 mddev->parallel_resync = n;
4511 if (mddev->sync_thread)
4512 wake_up(&resync_wait);
4517 /* force parallel resync, even with shared block devices */
4518 static struct md_sysfs_entry md_sync_force_parallel =
4519 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4520 sync_force_parallel_show, sync_force_parallel_store);
4523 sync_speed_show(struct mddev *mddev, char *page)
4525 unsigned long resync, dt, db;
4526 if (mddev->curr_resync == 0)
4527 return sprintf(page, "none\n");
4528 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4529 dt = (jiffies - mddev->resync_mark) / HZ;
4531 db = resync - mddev->resync_mark_cnt;
4532 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4535 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4538 sync_completed_show(struct mddev *mddev, char *page)
4540 unsigned long long max_sectors, resync;
4542 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4543 return sprintf(page, "none\n");
4545 if (mddev->curr_resync == 1 ||
4546 mddev->curr_resync == 2)
4547 return sprintf(page, "delayed\n");
4549 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4550 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4551 max_sectors = mddev->resync_max_sectors;
4553 max_sectors = mddev->dev_sectors;
4555 resync = mddev->curr_resync_completed;
4556 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4559 static struct md_sysfs_entry md_sync_completed =
4560 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4563 min_sync_show(struct mddev *mddev, char *page)
4565 return sprintf(page, "%llu\n",
4566 (unsigned long long)mddev->resync_min);
4569 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4571 unsigned long long min;
4574 if (kstrtoull(buf, 10, &min))
4577 spin_lock(&mddev->lock);
4579 if (min > mddev->resync_max)
4583 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4586 /* Round down to multiple of 4K for safety */
4587 mddev->resync_min = round_down(min, 8);
4591 spin_unlock(&mddev->lock);
4595 static struct md_sysfs_entry md_min_sync =
4596 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4599 max_sync_show(struct mddev *mddev, char *page)
4601 if (mddev->resync_max == MaxSector)
4602 return sprintf(page, "max\n");
4604 return sprintf(page, "%llu\n",
4605 (unsigned long long)mddev->resync_max);
4608 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4611 spin_lock(&mddev->lock);
4612 if (strncmp(buf, "max", 3) == 0)
4613 mddev->resync_max = MaxSector;
4615 unsigned long long max;
4619 if (kstrtoull(buf, 10, &max))
4621 if (max < mddev->resync_min)
4625 if (max < mddev->resync_max &&
4627 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4630 /* Must be a multiple of chunk_size */
4631 chunk = mddev->chunk_sectors;
4633 sector_t temp = max;
4636 if (sector_div(temp, chunk))
4639 mddev->resync_max = max;
4641 wake_up(&mddev->recovery_wait);
4644 spin_unlock(&mddev->lock);
4648 static struct md_sysfs_entry md_max_sync =
4649 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4652 suspend_lo_show(struct mddev *mddev, char *page)
4654 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4658 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4660 unsigned long long old, new;
4663 err = kstrtoull(buf, 10, &new);
4666 if (new != (sector_t)new)
4669 err = mddev_lock(mddev);
4673 if (mddev->pers == NULL ||
4674 mddev->pers->quiesce == NULL)
4676 old = mddev->suspend_lo;
4677 mddev->suspend_lo = new;
4679 /* Shrinking suspended region */
4680 mddev->pers->quiesce(mddev, 2);
4682 /* Expanding suspended region - need to wait */
4683 mddev->pers->quiesce(mddev, 1);
4684 mddev->pers->quiesce(mddev, 0);
4688 mddev_unlock(mddev);
4691 static struct md_sysfs_entry md_suspend_lo =
4692 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4695 suspend_hi_show(struct mddev *mddev, char *page)
4697 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4701 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4703 unsigned long long old, new;
4706 err = kstrtoull(buf, 10, &new);
4709 if (new != (sector_t)new)
4712 err = mddev_lock(mddev);
4716 if (mddev->pers == NULL ||
4717 mddev->pers->quiesce == NULL)
4719 old = mddev->suspend_hi;
4720 mddev->suspend_hi = new;
4722 /* Shrinking suspended region */
4723 mddev->pers->quiesce(mddev, 2);
4725 /* Expanding suspended region - need to wait */
4726 mddev->pers->quiesce(mddev, 1);
4727 mddev->pers->quiesce(mddev, 0);
4731 mddev_unlock(mddev);
4734 static struct md_sysfs_entry md_suspend_hi =
4735 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4738 reshape_position_show(struct mddev *mddev, char *page)
4740 if (mddev->reshape_position != MaxSector)
4741 return sprintf(page, "%llu\n",
4742 (unsigned long long)mddev->reshape_position);
4743 strcpy(page, "none\n");
4748 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4750 struct md_rdev *rdev;
4751 unsigned long long new;
4754 err = kstrtoull(buf, 10, &new);
4757 if (new != (sector_t)new)
4759 err = mddev_lock(mddev);
4765 mddev->reshape_position = new;
4766 mddev->delta_disks = 0;
4767 mddev->reshape_backwards = 0;
4768 mddev->new_level = mddev->level;
4769 mddev->new_layout = mddev->layout;
4770 mddev->new_chunk_sectors = mddev->chunk_sectors;
4771 rdev_for_each(rdev, mddev)
4772 rdev->new_data_offset = rdev->data_offset;
4775 mddev_unlock(mddev);
4779 static struct md_sysfs_entry md_reshape_position =
4780 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4781 reshape_position_store);
4784 reshape_direction_show(struct mddev *mddev, char *page)
4786 return sprintf(page, "%s\n",
4787 mddev->reshape_backwards ? "backwards" : "forwards");
4791 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4796 if (cmd_match(buf, "forwards"))
4798 else if (cmd_match(buf, "backwards"))
4802 if (mddev->reshape_backwards == backwards)
4805 err = mddev_lock(mddev);
4808 /* check if we are allowed to change */
4809 if (mddev->delta_disks)
4811 else if (mddev->persistent &&
4812 mddev->major_version == 0)
4815 mddev->reshape_backwards = backwards;
4816 mddev_unlock(mddev);
4820 static struct md_sysfs_entry md_reshape_direction =
4821 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4822 reshape_direction_store);
4825 array_size_show(struct mddev *mddev, char *page)
4827 if (mddev->external_size)
4828 return sprintf(page, "%llu\n",
4829 (unsigned long long)mddev->array_sectors/2);
4831 return sprintf(page, "default\n");
4835 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4840 err = mddev_lock(mddev);
4844 /* cluster raid doesn't support change array_sectors */
4845 if (mddev_is_clustered(mddev)) {
4846 mddev_unlock(mddev);
4850 if (strncmp(buf, "default", 7) == 0) {
4852 sectors = mddev->pers->size(mddev, 0, 0);
4854 sectors = mddev->array_sectors;
4856 mddev->external_size = 0;
4858 if (strict_blocks_to_sectors(buf, §ors) < 0)
4860 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4863 mddev->external_size = 1;
4867 mddev->array_sectors = sectors;
4869 set_capacity(mddev->gendisk, mddev->array_sectors);
4870 revalidate_disk(mddev->gendisk);
4873 mddev_unlock(mddev);
4877 static struct md_sysfs_entry md_array_size =
4878 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4881 static struct attribute *md_default_attrs[] = {
4884 &md_raid_disks.attr,
4885 &md_chunk_size.attr,
4887 &md_resync_start.attr,
4889 &md_new_device.attr,
4890 &md_safe_delay.attr,
4891 &md_array_state.attr,
4892 &md_reshape_position.attr,
4893 &md_reshape_direction.attr,
4894 &md_array_size.attr,
4895 &max_corr_read_errors.attr,
4899 static struct attribute *md_redundancy_attrs[] = {
4901 &md_last_scan_mode.attr,
4902 &md_mismatches.attr,
4905 &md_sync_speed.attr,
4906 &md_sync_force_parallel.attr,
4907 &md_sync_completed.attr,
4910 &md_suspend_lo.attr,
4911 &md_suspend_hi.attr,
4916 static struct attribute_group md_redundancy_group = {
4918 .attrs = md_redundancy_attrs,
4922 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4924 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4925 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4930 spin_lock(&all_mddevs_lock);
4931 if (list_empty(&mddev->all_mddevs)) {
4932 spin_unlock(&all_mddevs_lock);
4936 spin_unlock(&all_mddevs_lock);
4938 rv = entry->show(mddev, page);
4944 md_attr_store(struct kobject *kobj, struct attribute *attr,
4945 const char *page, size_t length)
4947 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4948 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4953 if (!capable(CAP_SYS_ADMIN))
4955 spin_lock(&all_mddevs_lock);
4956 if (list_empty(&mddev->all_mddevs)) {
4957 spin_unlock(&all_mddevs_lock);
4961 spin_unlock(&all_mddevs_lock);
4962 rv = entry->store(mddev, page, length);
4967 static void md_free(struct kobject *ko)
4969 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4971 if (mddev->sysfs_state)
4972 sysfs_put(mddev->sysfs_state);
4975 blk_cleanup_queue(mddev->queue);
4976 if (mddev->gendisk) {
4977 del_gendisk(mddev->gendisk);
4978 put_disk(mddev->gendisk);
4984 static const struct sysfs_ops md_sysfs_ops = {
4985 .show = md_attr_show,
4986 .store = md_attr_store,
4988 static struct kobj_type md_ktype = {
4990 .sysfs_ops = &md_sysfs_ops,
4991 .default_attrs = md_default_attrs,
4996 static void mddev_delayed_delete(struct work_struct *ws)
4998 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5000 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5001 kobject_del(&mddev->kobj);
5002 kobject_put(&mddev->kobj);
5005 static int md_alloc(dev_t dev, char *name)
5007 static DEFINE_MUTEX(disks_mutex);
5008 struct mddev *mddev = mddev_find(dev);
5009 struct gendisk *disk;
5018 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5019 shift = partitioned ? MdpMinorShift : 0;
5020 unit = MINOR(mddev->unit) >> shift;
5022 /* wait for any previous instance of this device to be
5023 * completely removed (mddev_delayed_delete).
5025 flush_workqueue(md_misc_wq);
5027 mutex_lock(&disks_mutex);
5033 /* Need to ensure that 'name' is not a duplicate.
5035 struct mddev *mddev2;
5036 spin_lock(&all_mddevs_lock);
5038 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5039 if (mddev2->gendisk &&
5040 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5041 spin_unlock(&all_mddevs_lock);
5044 spin_unlock(&all_mddevs_lock);
5048 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5051 mddev->queue->queuedata = mddev;
5053 blk_queue_make_request(mddev->queue, md_make_request);
5054 blk_set_stacking_limits(&mddev->queue->limits);
5056 disk = alloc_disk(1 << shift);
5058 blk_cleanup_queue(mddev->queue);
5059 mddev->queue = NULL;
5062 disk->major = MAJOR(mddev->unit);
5063 disk->first_minor = unit << shift;
5065 strcpy(disk->disk_name, name);
5066 else if (partitioned)
5067 sprintf(disk->disk_name, "md_d%d", unit);
5069 sprintf(disk->disk_name, "md%d", unit);
5070 disk->fops = &md_fops;
5071 disk->private_data = mddev;
5072 disk->queue = mddev->queue;
5073 blk_queue_write_cache(mddev->queue, true, true);
5074 /* Allow extended partitions. This makes the
5075 * 'mdp' device redundant, but we can't really
5078 disk->flags |= GENHD_FL_EXT_DEVT;
5079 mddev->gendisk = disk;
5080 /* As soon as we call add_disk(), another thread could get
5081 * through to md_open, so make sure it doesn't get too far
5083 mutex_lock(&mddev->open_mutex);
5086 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5087 &disk_to_dev(disk)->kobj, "%s", "md");
5089 /* This isn't possible, but as kobject_init_and_add is marked
5090 * __must_check, we must do something with the result
5092 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5096 if (mddev->kobj.sd &&
5097 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5098 printk(KERN_DEBUG "pointless warning\n");
5099 mutex_unlock(&mddev->open_mutex);
5101 mutex_unlock(&disks_mutex);
5102 if (!error && mddev->kobj.sd) {
5103 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5104 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5110 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5112 md_alloc(dev, NULL);
5116 static int add_named_array(const char *val, struct kernel_param *kp)
5118 /* val must be "md_*" where * is not all digits.
5119 * We allocate an array with a large free minor number, and
5120 * set the name to val. val must not already be an active name.
5122 int len = strlen(val);
5123 char buf[DISK_NAME_LEN];
5125 while (len && val[len-1] == '\n')
5127 if (len >= DISK_NAME_LEN)
5129 strlcpy(buf, val, len+1);
5130 if (strncmp(buf, "md_", 3) != 0)
5132 return md_alloc(0, buf);
5135 static void md_safemode_timeout(unsigned long data)
5137 struct mddev *mddev = (struct mddev *) data;
5139 if (!atomic_read(&mddev->writes_pending)) {
5140 mddev->safemode = 1;
5141 if (mddev->external)
5142 sysfs_notify_dirent_safe(mddev->sysfs_state);
5144 md_wakeup_thread(mddev->thread);
5147 static int start_dirty_degraded;
5149 int md_run(struct mddev *mddev)
5152 struct md_rdev *rdev;
5153 struct md_personality *pers;
5155 if (list_empty(&mddev->disks))
5156 /* cannot run an array with no devices.. */
5161 /* Cannot run until previous stop completes properly */
5162 if (mddev->sysfs_active)
5166 * Analyze all RAID superblock(s)
5168 if (!mddev->raid_disks) {
5169 if (!mddev->persistent)
5174 if (mddev->level != LEVEL_NONE)
5175 request_module("md-level-%d", mddev->level);
5176 else if (mddev->clevel[0])
5177 request_module("md-%s", mddev->clevel);
5180 * Drop all container device buffers, from now on
5181 * the only valid external interface is through the md
5184 rdev_for_each(rdev, mddev) {
5185 if (test_bit(Faulty, &rdev->flags))
5187 sync_blockdev(rdev->bdev);
5188 invalidate_bdev(rdev->bdev);
5190 /* perform some consistency tests on the device.
5191 * We don't want the data to overlap the metadata,
5192 * Internal Bitmap issues have been handled elsewhere.
5194 if (rdev->meta_bdev) {
5195 /* Nothing to check */;
5196 } else if (rdev->data_offset < rdev->sb_start) {
5197 if (mddev->dev_sectors &&
5198 rdev->data_offset + mddev->dev_sectors
5200 printk("md: %s: data overlaps metadata\n",
5205 if (rdev->sb_start + rdev->sb_size/512
5206 > rdev->data_offset) {
5207 printk("md: %s: metadata overlaps data\n",
5212 sysfs_notify_dirent_safe(rdev->sysfs_state);
5215 if (mddev->bio_set == NULL)
5216 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5218 spin_lock(&pers_lock);
5219 pers = find_pers(mddev->level, mddev->clevel);
5220 if (!pers || !try_module_get(pers->owner)) {
5221 spin_unlock(&pers_lock);
5222 if (mddev->level != LEVEL_NONE)
5223 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5226 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5230 spin_unlock(&pers_lock);
5231 if (mddev->level != pers->level) {
5232 mddev->level = pers->level;
5233 mddev->new_level = pers->level;
5235 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5237 if (mddev->reshape_position != MaxSector &&
5238 pers->start_reshape == NULL) {
5239 /* This personality cannot handle reshaping... */
5240 module_put(pers->owner);
5244 if (pers->sync_request) {
5245 /* Warn if this is a potentially silly
5248 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5249 struct md_rdev *rdev2;
5252 rdev_for_each(rdev, mddev)
5253 rdev_for_each(rdev2, mddev) {
5255 rdev->bdev->bd_contains ==
5256 rdev2->bdev->bd_contains) {
5258 "%s: WARNING: %s appears to be"
5259 " on the same physical disk as"
5262 bdevname(rdev->bdev,b),
5263 bdevname(rdev2->bdev,b2));
5270 "True protection against single-disk"
5271 " failure might be compromised.\n");
5274 mddev->recovery = 0;
5275 /* may be over-ridden by personality */
5276 mddev->resync_max_sectors = mddev->dev_sectors;
5278 mddev->ok_start_degraded = start_dirty_degraded;
5280 if (start_readonly && mddev->ro == 0)
5281 mddev->ro = 2; /* read-only, but switch on first write */
5283 err = pers->run(mddev);
5285 printk(KERN_ERR "md: pers->run() failed ...\n");
5286 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5287 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5288 " but 'external_size' not in effect?\n", __func__);
5290 "md: invalid array_size %llu > default size %llu\n",
5291 (unsigned long long)mddev->array_sectors / 2,
5292 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5295 if (err == 0 && pers->sync_request &&
5296 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5297 struct bitmap *bitmap;
5299 bitmap = bitmap_create(mddev, -1);
5300 if (IS_ERR(bitmap)) {
5301 err = PTR_ERR(bitmap);
5302 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5303 mdname(mddev), err);
5305 mddev->bitmap = bitmap;
5309 mddev_detach(mddev);
5311 pers->free(mddev, mddev->private);
5312 mddev->private = NULL;
5313 module_put(pers->owner);
5314 bitmap_destroy(mddev);
5320 rdev_for_each(rdev, mddev) {
5321 if (rdev->raid_disk >= 0 &&
5322 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5327 if (mddev->degraded)
5330 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5332 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5333 mddev->queue->backing_dev_info.congested_data = mddev;
5334 mddev->queue->backing_dev_info.congested_fn = md_congested;
5336 if (pers->sync_request) {
5337 if (mddev->kobj.sd &&
5338 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5340 "md: cannot register extra attributes for %s\n",
5342 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5343 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5346 atomic_set(&mddev->writes_pending,0);
5347 atomic_set(&mddev->max_corr_read_errors,
5348 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5349 mddev->safemode = 0;
5350 if (mddev_is_clustered(mddev))
5351 mddev->safemode_delay = 0;
5353 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5356 spin_lock(&mddev->lock);
5358 spin_unlock(&mddev->lock);
5359 rdev_for_each(rdev, mddev)
5360 if (rdev->raid_disk >= 0)
5361 if (sysfs_link_rdev(mddev, rdev))
5362 /* failure here is OK */;
5364 if (mddev->degraded && !mddev->ro)
5365 /* This ensures that recovering status is reported immediately
5366 * via sysfs - until a lack of spares is confirmed.
5368 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5369 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5371 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5372 md_update_sb(mddev, 0);
5374 md_new_event(mddev);
5375 sysfs_notify_dirent_safe(mddev->sysfs_state);
5376 sysfs_notify_dirent_safe(mddev->sysfs_action);
5377 sysfs_notify(&mddev->kobj, NULL, "degraded");
5380 EXPORT_SYMBOL_GPL(md_run);
5382 static int do_md_run(struct mddev *mddev)
5386 err = md_run(mddev);
5389 err = bitmap_load(mddev);
5391 bitmap_destroy(mddev);
5395 if (mddev_is_clustered(mddev))
5396 md_allow_write(mddev);
5398 md_wakeup_thread(mddev->thread);
5399 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5401 set_capacity(mddev->gendisk, mddev->array_sectors);
5402 revalidate_disk(mddev->gendisk);
5404 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5409 static int restart_array(struct mddev *mddev)
5411 struct gendisk *disk = mddev->gendisk;
5413 /* Complain if it has no devices */
5414 if (list_empty(&mddev->disks))
5420 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5421 struct md_rdev *rdev;
5422 bool has_journal = false;
5425 rdev_for_each_rcu(rdev, mddev) {
5426 if (test_bit(Journal, &rdev->flags) &&
5427 !test_bit(Faulty, &rdev->flags)) {
5434 /* Don't restart rw with journal missing/faulty */
5439 mddev->safemode = 0;
5441 set_disk_ro(disk, 0);
5442 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5444 /* Kick recovery or resync if necessary */
5445 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5446 md_wakeup_thread(mddev->thread);
5447 md_wakeup_thread(mddev->sync_thread);
5448 sysfs_notify_dirent_safe(mddev->sysfs_state);
5452 static void md_clean(struct mddev *mddev)
5454 mddev->array_sectors = 0;
5455 mddev->external_size = 0;
5456 mddev->dev_sectors = 0;
5457 mddev->raid_disks = 0;
5458 mddev->recovery_cp = 0;
5459 mddev->resync_min = 0;
5460 mddev->resync_max = MaxSector;
5461 mddev->reshape_position = MaxSector;
5462 mddev->external = 0;
5463 mddev->persistent = 0;
5464 mddev->level = LEVEL_NONE;
5465 mddev->clevel[0] = 0;
5468 mddev->metadata_type[0] = 0;
5469 mddev->chunk_sectors = 0;
5470 mddev->ctime = mddev->utime = 0;
5472 mddev->max_disks = 0;
5474 mddev->can_decrease_events = 0;
5475 mddev->delta_disks = 0;
5476 mddev->reshape_backwards = 0;
5477 mddev->new_level = LEVEL_NONE;
5478 mddev->new_layout = 0;
5479 mddev->new_chunk_sectors = 0;
5480 mddev->curr_resync = 0;
5481 atomic64_set(&mddev->resync_mismatches, 0);
5482 mddev->suspend_lo = mddev->suspend_hi = 0;
5483 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5484 mddev->recovery = 0;
5487 mddev->degraded = 0;
5488 mddev->safemode = 0;
5489 mddev->private = NULL;
5490 mddev->cluster_info = NULL;
5491 mddev->bitmap_info.offset = 0;
5492 mddev->bitmap_info.default_offset = 0;
5493 mddev->bitmap_info.default_space = 0;
5494 mddev->bitmap_info.chunksize = 0;
5495 mddev->bitmap_info.daemon_sleep = 0;
5496 mddev->bitmap_info.max_write_behind = 0;
5497 mddev->bitmap_info.nodes = 0;
5500 static void __md_stop_writes(struct mddev *mddev)
5502 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5503 flush_workqueue(md_misc_wq);
5504 if (mddev->sync_thread) {
5505 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5506 md_reap_sync_thread(mddev);
5509 del_timer_sync(&mddev->safemode_timer);
5511 bitmap_flush(mddev);
5512 md_super_wait(mddev);
5514 if (mddev->ro == 0 &&
5515 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5516 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5517 /* mark array as shutdown cleanly */
5518 if (!mddev_is_clustered(mddev))
5520 md_update_sb(mddev, 1);
5524 void md_stop_writes(struct mddev *mddev)
5526 mddev_lock_nointr(mddev);
5527 __md_stop_writes(mddev);
5528 mddev_unlock(mddev);
5530 EXPORT_SYMBOL_GPL(md_stop_writes);
5532 static void mddev_detach(struct mddev *mddev)
5534 struct bitmap *bitmap = mddev->bitmap;
5535 /* wait for behind writes to complete */
5536 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5537 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5539 /* need to kick something here to make sure I/O goes? */
5540 wait_event(bitmap->behind_wait,
5541 atomic_read(&bitmap->behind_writes) == 0);
5543 if (mddev->pers && mddev->pers->quiesce) {
5544 mddev->pers->quiesce(mddev, 1);
5545 mddev->pers->quiesce(mddev, 0);
5547 md_unregister_thread(&mddev->thread);
5549 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5552 static void __md_stop(struct mddev *mddev)
5554 struct md_personality *pers = mddev->pers;
5555 mddev_detach(mddev);
5556 /* Ensure ->event_work is done */
5557 flush_workqueue(md_misc_wq);
5558 spin_lock(&mddev->lock);
5560 spin_unlock(&mddev->lock);
5561 pers->free(mddev, mddev->private);
5562 mddev->private = NULL;
5563 if (pers->sync_request && mddev->to_remove == NULL)
5564 mddev->to_remove = &md_redundancy_group;
5565 module_put(pers->owner);
5566 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5569 void md_stop(struct mddev *mddev)
5571 /* stop the array and free an attached data structures.
5572 * This is called from dm-raid
5575 bitmap_destroy(mddev);
5577 bioset_free(mddev->bio_set);
5580 EXPORT_SYMBOL_GPL(md_stop);
5582 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5587 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5589 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5590 md_wakeup_thread(mddev->thread);
5592 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5593 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5594 if (mddev->sync_thread)
5595 /* Thread might be blocked waiting for metadata update
5596 * which will now never happen */
5597 wake_up_process(mddev->sync_thread->tsk);
5599 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5601 mddev_unlock(mddev);
5602 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5604 wait_event(mddev->sb_wait,
5605 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5606 mddev_lock_nointr(mddev);
5608 mutex_lock(&mddev->open_mutex);
5609 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5610 mddev->sync_thread ||
5611 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5612 printk("md: %s still in use.\n",mdname(mddev));
5614 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5615 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5616 md_wakeup_thread(mddev->thread);
5622 __md_stop_writes(mddev);
5628 set_disk_ro(mddev->gendisk, 1);
5629 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5630 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5631 md_wakeup_thread(mddev->thread);
5632 sysfs_notify_dirent_safe(mddev->sysfs_state);
5636 mutex_unlock(&mddev->open_mutex);
5641 * 0 - completely stop and dis-assemble array
5642 * 2 - stop but do not disassemble array
5644 static int do_md_stop(struct mddev *mddev, int mode,
5645 struct block_device *bdev)
5647 struct gendisk *disk = mddev->gendisk;
5648 struct md_rdev *rdev;
5651 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5653 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5654 md_wakeup_thread(mddev->thread);
5656 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5657 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5658 if (mddev->sync_thread)
5659 /* Thread might be blocked waiting for metadata update
5660 * which will now never happen */
5661 wake_up_process(mddev->sync_thread->tsk);
5663 mddev_unlock(mddev);
5664 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5665 !test_bit(MD_RECOVERY_RUNNING,
5666 &mddev->recovery)));
5667 mddev_lock_nointr(mddev);
5669 mutex_lock(&mddev->open_mutex);
5670 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5671 mddev->sysfs_active ||
5672 mddev->sync_thread ||
5673 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5674 printk("md: %s still in use.\n",mdname(mddev));
5675 mutex_unlock(&mddev->open_mutex);
5677 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5678 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5679 md_wakeup_thread(mddev->thread);
5685 set_disk_ro(disk, 0);
5687 __md_stop_writes(mddev);
5689 mddev->queue->backing_dev_info.congested_fn = NULL;
5691 /* tell userspace to handle 'inactive' */
5692 sysfs_notify_dirent_safe(mddev->sysfs_state);
5694 rdev_for_each(rdev, mddev)
5695 if (rdev->raid_disk >= 0)
5696 sysfs_unlink_rdev(mddev, rdev);
5698 set_capacity(disk, 0);
5699 mutex_unlock(&mddev->open_mutex);
5701 revalidate_disk(disk);
5706 mutex_unlock(&mddev->open_mutex);
5708 * Free resources if final stop
5711 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5713 bitmap_destroy(mddev);
5714 if (mddev->bitmap_info.file) {
5715 struct file *f = mddev->bitmap_info.file;
5716 spin_lock(&mddev->lock);
5717 mddev->bitmap_info.file = NULL;
5718 spin_unlock(&mddev->lock);
5721 mddev->bitmap_info.offset = 0;
5723 export_array(mddev);
5726 if (mddev->hold_active == UNTIL_STOP)
5727 mddev->hold_active = 0;
5729 md_new_event(mddev);
5730 sysfs_notify_dirent_safe(mddev->sysfs_state);
5735 static void autorun_array(struct mddev *mddev)
5737 struct md_rdev *rdev;
5740 if (list_empty(&mddev->disks))
5743 printk(KERN_INFO "md: running: ");
5745 rdev_for_each(rdev, mddev) {
5746 char b[BDEVNAME_SIZE];
5747 printk("<%s>", bdevname(rdev->bdev,b));
5751 err = do_md_run(mddev);
5753 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5754 do_md_stop(mddev, 0, NULL);
5759 * lets try to run arrays based on all disks that have arrived
5760 * until now. (those are in pending_raid_disks)
5762 * the method: pick the first pending disk, collect all disks with
5763 * the same UUID, remove all from the pending list and put them into
5764 * the 'same_array' list. Then order this list based on superblock
5765 * update time (freshest comes first), kick out 'old' disks and
5766 * compare superblocks. If everything's fine then run it.
5768 * If "unit" is allocated, then bump its reference count
5770 static void autorun_devices(int part)
5772 struct md_rdev *rdev0, *rdev, *tmp;
5773 struct mddev *mddev;
5774 char b[BDEVNAME_SIZE];
5776 printk(KERN_INFO "md: autorun ...\n");
5777 while (!list_empty(&pending_raid_disks)) {
5780 LIST_HEAD(candidates);
5781 rdev0 = list_entry(pending_raid_disks.next,
5782 struct md_rdev, same_set);
5784 printk(KERN_INFO "md: considering %s ...\n",
5785 bdevname(rdev0->bdev,b));
5786 INIT_LIST_HEAD(&candidates);
5787 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5788 if (super_90_load(rdev, rdev0, 0) >= 0) {
5789 printk(KERN_INFO "md: adding %s ...\n",
5790 bdevname(rdev->bdev,b));
5791 list_move(&rdev->same_set, &candidates);
5794 * now we have a set of devices, with all of them having
5795 * mostly sane superblocks. It's time to allocate the
5799 dev = MKDEV(mdp_major,
5800 rdev0->preferred_minor << MdpMinorShift);
5801 unit = MINOR(dev) >> MdpMinorShift;
5803 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5806 if (rdev0->preferred_minor != unit) {
5807 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5808 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5812 md_probe(dev, NULL, NULL);
5813 mddev = mddev_find(dev);
5814 if (!mddev || !mddev->gendisk) {
5818 "md: cannot allocate memory for md drive.\n");
5821 if (mddev_lock(mddev))
5822 printk(KERN_WARNING "md: %s locked, cannot run\n",
5824 else if (mddev->raid_disks || mddev->major_version
5825 || !list_empty(&mddev->disks)) {
5827 "md: %s already running, cannot run %s\n",
5828 mdname(mddev), bdevname(rdev0->bdev,b));
5829 mddev_unlock(mddev);
5831 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5832 mddev->persistent = 1;
5833 rdev_for_each_list(rdev, tmp, &candidates) {
5834 list_del_init(&rdev->same_set);
5835 if (bind_rdev_to_array(rdev, mddev))
5838 autorun_array(mddev);
5839 mddev_unlock(mddev);
5841 /* on success, candidates will be empty, on error
5844 rdev_for_each_list(rdev, tmp, &candidates) {
5845 list_del_init(&rdev->same_set);
5850 printk(KERN_INFO "md: ... autorun DONE.\n");
5852 #endif /* !MODULE */
5854 static int get_version(void __user *arg)
5858 ver.major = MD_MAJOR_VERSION;
5859 ver.minor = MD_MINOR_VERSION;
5860 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5862 if (copy_to_user(arg, &ver, sizeof(ver)))
5868 static int get_array_info(struct mddev *mddev, void __user *arg)
5870 mdu_array_info_t info;
5871 int nr,working,insync,failed,spare;
5872 struct md_rdev *rdev;
5874 nr = working = insync = failed = spare = 0;
5876 rdev_for_each_rcu(rdev, mddev) {
5878 if (test_bit(Faulty, &rdev->flags))
5882 if (test_bit(In_sync, &rdev->flags))
5884 else if (test_bit(Journal, &rdev->flags))
5885 /* TODO: add journal count to md_u.h */
5893 info.major_version = mddev->major_version;
5894 info.minor_version = mddev->minor_version;
5895 info.patch_version = MD_PATCHLEVEL_VERSION;
5896 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5897 info.level = mddev->level;
5898 info.size = mddev->dev_sectors / 2;
5899 if (info.size != mddev->dev_sectors / 2) /* overflow */
5902 info.raid_disks = mddev->raid_disks;
5903 info.md_minor = mddev->md_minor;
5904 info.not_persistent= !mddev->persistent;
5906 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5909 info.state = (1<<MD_SB_CLEAN);
5910 if (mddev->bitmap && mddev->bitmap_info.offset)
5911 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5912 if (mddev_is_clustered(mddev))
5913 info.state |= (1<<MD_SB_CLUSTERED);
5914 info.active_disks = insync;
5915 info.working_disks = working;
5916 info.failed_disks = failed;
5917 info.spare_disks = spare;
5919 info.layout = mddev->layout;
5920 info.chunk_size = mddev->chunk_sectors << 9;
5922 if (copy_to_user(arg, &info, sizeof(info)))
5928 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5930 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5934 file = kzalloc(sizeof(*file), GFP_NOIO);
5939 spin_lock(&mddev->lock);
5940 /* bitmap enabled */
5941 if (mddev->bitmap_info.file) {
5942 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5943 sizeof(file->pathname));
5947 memmove(file->pathname, ptr,
5948 sizeof(file->pathname)-(ptr-file->pathname));
5950 spin_unlock(&mddev->lock);
5953 copy_to_user(arg, file, sizeof(*file)))
5960 static int get_disk_info(struct mddev *mddev, void __user * arg)
5962 mdu_disk_info_t info;
5963 struct md_rdev *rdev;
5965 if (copy_from_user(&info, arg, sizeof(info)))
5969 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5971 info.major = MAJOR(rdev->bdev->bd_dev);
5972 info.minor = MINOR(rdev->bdev->bd_dev);
5973 info.raid_disk = rdev->raid_disk;
5975 if (test_bit(Faulty, &rdev->flags))
5976 info.state |= (1<<MD_DISK_FAULTY);
5977 else if (test_bit(In_sync, &rdev->flags)) {
5978 info.state |= (1<<MD_DISK_ACTIVE);
5979 info.state |= (1<<MD_DISK_SYNC);
5981 if (test_bit(Journal, &rdev->flags))
5982 info.state |= (1<<MD_DISK_JOURNAL);
5983 if (test_bit(WriteMostly, &rdev->flags))
5984 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5986 info.major = info.minor = 0;
5987 info.raid_disk = -1;
5988 info.state = (1<<MD_DISK_REMOVED);
5992 if (copy_to_user(arg, &info, sizeof(info)))
5998 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6000 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6001 struct md_rdev *rdev;
6002 dev_t dev = MKDEV(info->major,info->minor);
6004 if (mddev_is_clustered(mddev) &&
6005 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6006 pr_err("%s: Cannot add to clustered mddev.\n",
6011 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6014 if (!mddev->raid_disks) {
6016 /* expecting a device which has a superblock */
6017 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6020 "md: md_import_device returned %ld\n",
6022 return PTR_ERR(rdev);
6024 if (!list_empty(&mddev->disks)) {
6025 struct md_rdev *rdev0
6026 = list_entry(mddev->disks.next,
6027 struct md_rdev, same_set);
6028 err = super_types[mddev->major_version]
6029 .load_super(rdev, rdev0, mddev->minor_version);
6032 "md: %s has different UUID to %s\n",
6033 bdevname(rdev->bdev,b),
6034 bdevname(rdev0->bdev,b2));
6039 err = bind_rdev_to_array(rdev, mddev);
6046 * add_new_disk can be used once the array is assembled
6047 * to add "hot spares". They must already have a superblock
6052 if (!mddev->pers->hot_add_disk) {
6054 "%s: personality does not support diskops!\n",
6058 if (mddev->persistent)
6059 rdev = md_import_device(dev, mddev->major_version,
6060 mddev->minor_version);
6062 rdev = md_import_device(dev, -1, -1);
6065 "md: md_import_device returned %ld\n",
6067 return PTR_ERR(rdev);
6069 /* set saved_raid_disk if appropriate */
6070 if (!mddev->persistent) {
6071 if (info->state & (1<<MD_DISK_SYNC) &&
6072 info->raid_disk < mddev->raid_disks) {
6073 rdev->raid_disk = info->raid_disk;
6074 set_bit(In_sync, &rdev->flags);
6075 clear_bit(Bitmap_sync, &rdev->flags);
6077 rdev->raid_disk = -1;
6078 rdev->saved_raid_disk = rdev->raid_disk;
6080 super_types[mddev->major_version].
6081 validate_super(mddev, rdev);
6082 if ((info->state & (1<<MD_DISK_SYNC)) &&
6083 rdev->raid_disk != info->raid_disk) {
6084 /* This was a hot-add request, but events doesn't
6085 * match, so reject it.
6091 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6092 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6093 set_bit(WriteMostly, &rdev->flags);
6095 clear_bit(WriteMostly, &rdev->flags);
6097 if (info->state & (1<<MD_DISK_JOURNAL)) {
6098 struct md_rdev *rdev2;
6099 bool has_journal = false;
6101 /* make sure no existing journal disk */
6102 rdev_for_each(rdev2, mddev) {
6103 if (test_bit(Journal, &rdev2->flags)) {
6112 set_bit(Journal, &rdev->flags);
6115 * check whether the device shows up in other nodes
6117 if (mddev_is_clustered(mddev)) {
6118 if (info->state & (1 << MD_DISK_CANDIDATE))
6119 set_bit(Candidate, &rdev->flags);
6120 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6121 /* --add initiated by this node */
6122 err = md_cluster_ops->add_new_disk(mddev, rdev);
6130 rdev->raid_disk = -1;
6131 err = bind_rdev_to_array(rdev, mddev);
6136 if (mddev_is_clustered(mddev)) {
6137 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6139 err = md_cluster_ops->new_disk_ack(mddev,
6142 md_kick_rdev_from_array(rdev);
6146 md_cluster_ops->add_new_disk_cancel(mddev);
6148 err = add_bound_rdev(rdev);
6152 err = add_bound_rdev(rdev);
6157 /* otherwise, add_new_disk is only allowed
6158 * for major_version==0 superblocks
6160 if (mddev->major_version != 0) {
6161 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6166 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6168 rdev = md_import_device(dev, -1, 0);
6171 "md: error, md_import_device() returned %ld\n",
6173 return PTR_ERR(rdev);
6175 rdev->desc_nr = info->number;
6176 if (info->raid_disk < mddev->raid_disks)
6177 rdev->raid_disk = info->raid_disk;
6179 rdev->raid_disk = -1;
6181 if (rdev->raid_disk < mddev->raid_disks)
6182 if (info->state & (1<<MD_DISK_SYNC))
6183 set_bit(In_sync, &rdev->flags);
6185 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6186 set_bit(WriteMostly, &rdev->flags);
6188 if (!mddev->persistent) {
6189 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6190 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6192 rdev->sb_start = calc_dev_sboffset(rdev);
6193 rdev->sectors = rdev->sb_start;
6195 err = bind_rdev_to_array(rdev, mddev);
6205 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6207 char b[BDEVNAME_SIZE];
6208 struct md_rdev *rdev;
6213 rdev = find_rdev(mddev, dev);
6217 if (rdev->raid_disk < 0)
6220 clear_bit(Blocked, &rdev->flags);
6221 remove_and_add_spares(mddev, rdev);
6223 if (rdev->raid_disk >= 0)
6227 if (mddev_is_clustered(mddev))
6228 md_cluster_ops->remove_disk(mddev, rdev);
6230 md_kick_rdev_from_array(rdev);
6231 md_update_sb(mddev, 1);
6232 md_new_event(mddev);
6236 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6237 bdevname(rdev->bdev,b), mdname(mddev));
6241 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6243 char b[BDEVNAME_SIZE];
6245 struct md_rdev *rdev;
6250 if (mddev->major_version != 0) {
6251 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6252 " version-0 superblocks.\n",
6256 if (!mddev->pers->hot_add_disk) {
6258 "%s: personality does not support diskops!\n",
6263 rdev = md_import_device(dev, -1, 0);
6266 "md: error, md_import_device() returned %ld\n",
6271 if (mddev->persistent)
6272 rdev->sb_start = calc_dev_sboffset(rdev);
6274 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6276 rdev->sectors = rdev->sb_start;
6278 if (test_bit(Faulty, &rdev->flags)) {
6280 "md: can not hot-add faulty %s disk to %s!\n",
6281 bdevname(rdev->bdev,b), mdname(mddev));
6286 clear_bit(In_sync, &rdev->flags);
6288 rdev->saved_raid_disk = -1;
6289 err = bind_rdev_to_array(rdev, mddev);
6294 * The rest should better be atomic, we can have disk failures
6295 * noticed in interrupt contexts ...
6298 rdev->raid_disk = -1;
6300 md_update_sb(mddev, 1);
6302 * Kick recovery, maybe this spare has to be added to the
6303 * array immediately.
6305 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6306 md_wakeup_thread(mddev->thread);
6307 md_new_event(mddev);
6315 static int set_bitmap_file(struct mddev *mddev, int fd)
6320 if (!mddev->pers->quiesce || !mddev->thread)
6322 if (mddev->recovery || mddev->sync_thread)
6324 /* we should be able to change the bitmap.. */
6328 struct inode *inode;
6331 if (mddev->bitmap || mddev->bitmap_info.file)
6332 return -EEXIST; /* cannot add when bitmap is present */
6336 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6341 inode = f->f_mapping->host;
6342 if (!S_ISREG(inode->i_mode)) {
6343 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6346 } else if (!(f->f_mode & FMODE_WRITE)) {
6347 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6350 } else if (atomic_read(&inode->i_writecount) != 1) {
6351 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6359 mddev->bitmap_info.file = f;
6360 mddev->bitmap_info.offset = 0; /* file overrides offset */
6361 } else if (mddev->bitmap == NULL)
6362 return -ENOENT; /* cannot remove what isn't there */
6365 mddev->pers->quiesce(mddev, 1);
6367 struct bitmap *bitmap;
6369 bitmap = bitmap_create(mddev, -1);
6370 if (!IS_ERR(bitmap)) {
6371 mddev->bitmap = bitmap;
6372 err = bitmap_load(mddev);
6374 err = PTR_ERR(bitmap);
6376 if (fd < 0 || err) {
6377 bitmap_destroy(mddev);
6378 fd = -1; /* make sure to put the file */
6380 mddev->pers->quiesce(mddev, 0);
6383 struct file *f = mddev->bitmap_info.file;
6385 spin_lock(&mddev->lock);
6386 mddev->bitmap_info.file = NULL;
6387 spin_unlock(&mddev->lock);
6396 * set_array_info is used two different ways
6397 * The original usage is when creating a new array.
6398 * In this usage, raid_disks is > 0 and it together with
6399 * level, size, not_persistent,layout,chunksize determine the
6400 * shape of the array.
6401 * This will always create an array with a type-0.90.0 superblock.
6402 * The newer usage is when assembling an array.
6403 * In this case raid_disks will be 0, and the major_version field is
6404 * use to determine which style super-blocks are to be found on the devices.
6405 * The minor and patch _version numbers are also kept incase the
6406 * super_block handler wishes to interpret them.
6408 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6411 if (info->raid_disks == 0) {
6412 /* just setting version number for superblock loading */
6413 if (info->major_version < 0 ||
6414 info->major_version >= ARRAY_SIZE(super_types) ||
6415 super_types[info->major_version].name == NULL) {
6416 /* maybe try to auto-load a module? */
6418 "md: superblock version %d not known\n",
6419 info->major_version);
6422 mddev->major_version = info->major_version;
6423 mddev->minor_version = info->minor_version;
6424 mddev->patch_version = info->patch_version;
6425 mddev->persistent = !info->not_persistent;
6426 /* ensure mddev_put doesn't delete this now that there
6427 * is some minimal configuration.
6429 mddev->ctime = ktime_get_real_seconds();
6432 mddev->major_version = MD_MAJOR_VERSION;
6433 mddev->minor_version = MD_MINOR_VERSION;
6434 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6435 mddev->ctime = ktime_get_real_seconds();
6437 mddev->level = info->level;
6438 mddev->clevel[0] = 0;
6439 mddev->dev_sectors = 2 * (sector_t)info->size;
6440 mddev->raid_disks = info->raid_disks;
6441 /* don't set md_minor, it is determined by which /dev/md* was
6444 if (info->state & (1<<MD_SB_CLEAN))
6445 mddev->recovery_cp = MaxSector;
6447 mddev->recovery_cp = 0;
6448 mddev->persistent = ! info->not_persistent;
6449 mddev->external = 0;
6451 mddev->layout = info->layout;
6452 mddev->chunk_sectors = info->chunk_size >> 9;
6454 mddev->max_disks = MD_SB_DISKS;
6456 if (mddev->persistent)
6458 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6460 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6461 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6462 mddev->bitmap_info.offset = 0;
6464 mddev->reshape_position = MaxSector;
6467 * Generate a 128 bit UUID
6469 get_random_bytes(mddev->uuid, 16);
6471 mddev->new_level = mddev->level;
6472 mddev->new_chunk_sectors = mddev->chunk_sectors;
6473 mddev->new_layout = mddev->layout;
6474 mddev->delta_disks = 0;
6475 mddev->reshape_backwards = 0;
6480 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6482 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6484 if (mddev->external_size)
6487 mddev->array_sectors = array_sectors;
6489 EXPORT_SYMBOL(md_set_array_sectors);
6491 static int update_size(struct mddev *mddev, sector_t num_sectors)
6493 struct md_rdev *rdev;
6495 int fit = (num_sectors == 0);
6497 /* cluster raid doesn't support update size */
6498 if (mddev_is_clustered(mddev))
6501 if (mddev->pers->resize == NULL)
6503 /* The "num_sectors" is the number of sectors of each device that
6504 * is used. This can only make sense for arrays with redundancy.
6505 * linear and raid0 always use whatever space is available. We can only
6506 * consider changing this number if no resync or reconstruction is
6507 * happening, and if the new size is acceptable. It must fit before the
6508 * sb_start or, if that is <data_offset, it must fit before the size
6509 * of each device. If num_sectors is zero, we find the largest size
6512 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6518 rdev_for_each(rdev, mddev) {
6519 sector_t avail = rdev->sectors;
6521 if (fit && (num_sectors == 0 || num_sectors > avail))
6522 num_sectors = avail;
6523 if (avail < num_sectors)
6526 rv = mddev->pers->resize(mddev, num_sectors);
6528 revalidate_disk(mddev->gendisk);
6532 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6535 struct md_rdev *rdev;
6536 /* change the number of raid disks */
6537 if (mddev->pers->check_reshape == NULL)
6541 if (raid_disks <= 0 ||
6542 (mddev->max_disks && raid_disks >= mddev->max_disks))
6544 if (mddev->sync_thread ||
6545 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6546 mddev->reshape_position != MaxSector)
6549 rdev_for_each(rdev, mddev) {
6550 if (mddev->raid_disks < raid_disks &&
6551 rdev->data_offset < rdev->new_data_offset)
6553 if (mddev->raid_disks > raid_disks &&
6554 rdev->data_offset > rdev->new_data_offset)
6558 mddev->delta_disks = raid_disks - mddev->raid_disks;
6559 if (mddev->delta_disks < 0)
6560 mddev->reshape_backwards = 1;
6561 else if (mddev->delta_disks > 0)
6562 mddev->reshape_backwards = 0;
6564 rv = mddev->pers->check_reshape(mddev);
6566 mddev->delta_disks = 0;
6567 mddev->reshape_backwards = 0;
6573 * update_array_info is used to change the configuration of an
6575 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6576 * fields in the info are checked against the array.
6577 * Any differences that cannot be handled will cause an error.
6578 * Normally, only one change can be managed at a time.
6580 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6586 /* calculate expected state,ignoring low bits */
6587 if (mddev->bitmap && mddev->bitmap_info.offset)
6588 state |= (1 << MD_SB_BITMAP_PRESENT);
6590 if (mddev->major_version != info->major_version ||
6591 mddev->minor_version != info->minor_version ||
6592 /* mddev->patch_version != info->patch_version || */
6593 mddev->ctime != info->ctime ||
6594 mddev->level != info->level ||
6595 /* mddev->layout != info->layout || */
6596 mddev->persistent != !info->not_persistent ||
6597 mddev->chunk_sectors != info->chunk_size >> 9 ||
6598 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6599 ((state^info->state) & 0xfffffe00)
6602 /* Check there is only one change */
6603 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6605 if (mddev->raid_disks != info->raid_disks)
6607 if (mddev->layout != info->layout)
6609 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6616 if (mddev->layout != info->layout) {
6618 * we don't need to do anything at the md level, the
6619 * personality will take care of it all.
6621 if (mddev->pers->check_reshape == NULL)
6624 mddev->new_layout = info->layout;
6625 rv = mddev->pers->check_reshape(mddev);
6627 mddev->new_layout = mddev->layout;
6631 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6632 rv = update_size(mddev, (sector_t)info->size * 2);
6634 if (mddev->raid_disks != info->raid_disks)
6635 rv = update_raid_disks(mddev, info->raid_disks);
6637 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6638 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6642 if (mddev->recovery || mddev->sync_thread) {
6646 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6647 struct bitmap *bitmap;
6648 /* add the bitmap */
6649 if (mddev->bitmap) {
6653 if (mddev->bitmap_info.default_offset == 0) {
6657 mddev->bitmap_info.offset =
6658 mddev->bitmap_info.default_offset;
6659 mddev->bitmap_info.space =
6660 mddev->bitmap_info.default_space;
6661 mddev->pers->quiesce(mddev, 1);
6662 bitmap = bitmap_create(mddev, -1);
6663 if (!IS_ERR(bitmap)) {
6664 mddev->bitmap = bitmap;
6665 rv = bitmap_load(mddev);
6667 rv = PTR_ERR(bitmap);
6669 bitmap_destroy(mddev);
6670 mddev->pers->quiesce(mddev, 0);
6672 /* remove the bitmap */
6673 if (!mddev->bitmap) {
6677 if (mddev->bitmap->storage.file) {
6681 if (mddev->bitmap_info.nodes) {
6682 /* hold PW on all the bitmap lock */
6683 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6684 printk("md: can't change bitmap to none since the"
6685 " array is in use by more than one node\n");
6687 md_cluster_ops->unlock_all_bitmaps(mddev);
6691 mddev->bitmap_info.nodes = 0;
6692 md_cluster_ops->leave(mddev);
6694 mddev->pers->quiesce(mddev, 1);
6695 bitmap_destroy(mddev);
6696 mddev->pers->quiesce(mddev, 0);
6697 mddev->bitmap_info.offset = 0;
6700 md_update_sb(mddev, 1);
6706 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6708 struct md_rdev *rdev;
6711 if (mddev->pers == NULL)
6715 rdev = find_rdev_rcu(mddev, dev);
6719 md_error(mddev, rdev);
6720 if (!test_bit(Faulty, &rdev->flags))
6728 * We have a problem here : there is no easy way to give a CHS
6729 * virtual geometry. We currently pretend that we have a 2 heads
6730 * 4 sectors (with a BIG number of cylinders...). This drives
6731 * dosfs just mad... ;-)
6733 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6735 struct mddev *mddev = bdev->bd_disk->private_data;
6739 geo->cylinders = mddev->array_sectors / 8;
6743 static inline bool md_ioctl_valid(unsigned int cmd)
6748 case GET_ARRAY_INFO:
6749 case GET_BITMAP_FILE:
6752 case HOT_REMOVE_DISK:
6755 case RESTART_ARRAY_RW:
6757 case SET_ARRAY_INFO:
6758 case SET_BITMAP_FILE:
6759 case SET_DISK_FAULTY:
6762 case CLUSTERED_DISK_NACK:
6769 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6770 unsigned int cmd, unsigned long arg)
6773 void __user *argp = (void __user *)arg;
6774 struct mddev *mddev = NULL;
6776 bool did_set_md_closing = false;
6778 if (!md_ioctl_valid(cmd))
6783 case GET_ARRAY_INFO:
6787 if (!capable(CAP_SYS_ADMIN))
6792 * Commands dealing with the RAID driver but not any
6797 err = get_version(argp);
6803 autostart_arrays(arg);
6810 * Commands creating/starting a new array:
6813 mddev = bdev->bd_disk->private_data;
6820 /* Some actions do not requires the mutex */
6822 case GET_ARRAY_INFO:
6823 if (!mddev->raid_disks && !mddev->external)
6826 err = get_array_info(mddev, argp);
6830 if (!mddev->raid_disks && !mddev->external)
6833 err = get_disk_info(mddev, argp);
6836 case SET_DISK_FAULTY:
6837 err = set_disk_faulty(mddev, new_decode_dev(arg));
6840 case GET_BITMAP_FILE:
6841 err = get_bitmap_file(mddev, argp);
6846 if (cmd == ADD_NEW_DISK)
6847 /* need to ensure md_delayed_delete() has completed */
6848 flush_workqueue(md_misc_wq);
6850 if (cmd == HOT_REMOVE_DISK)
6851 /* need to ensure recovery thread has run */
6852 wait_event_interruptible_timeout(mddev->sb_wait,
6853 !test_bit(MD_RECOVERY_NEEDED,
6855 msecs_to_jiffies(5000));
6856 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6857 /* Need to flush page cache, and ensure no-one else opens
6860 mutex_lock(&mddev->open_mutex);
6861 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6862 mutex_unlock(&mddev->open_mutex);
6866 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
6867 mutex_unlock(&mddev->open_mutex);
6871 did_set_md_closing = true;
6872 mutex_unlock(&mddev->open_mutex);
6873 sync_blockdev(bdev);
6875 err = mddev_lock(mddev);
6878 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6883 if (cmd == SET_ARRAY_INFO) {
6884 mdu_array_info_t info;
6886 memset(&info, 0, sizeof(info));
6887 else if (copy_from_user(&info, argp, sizeof(info))) {
6892 err = update_array_info(mddev, &info);
6894 printk(KERN_WARNING "md: couldn't update"
6895 " array info. %d\n", err);
6900 if (!list_empty(&mddev->disks)) {
6902 "md: array %s already has disks!\n",
6907 if (mddev->raid_disks) {
6909 "md: array %s already initialised!\n",
6914 err = set_array_info(mddev, &info);
6916 printk(KERN_WARNING "md: couldn't set"
6917 " array info. %d\n", err);
6924 * Commands querying/configuring an existing array:
6926 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6927 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6928 if ((!mddev->raid_disks && !mddev->external)
6929 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6930 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6931 && cmd != GET_BITMAP_FILE) {
6937 * Commands even a read-only array can execute:
6940 case RESTART_ARRAY_RW:
6941 err = restart_array(mddev);
6945 err = do_md_stop(mddev, 0, bdev);
6949 err = md_set_readonly(mddev, bdev);
6952 case HOT_REMOVE_DISK:
6953 err = hot_remove_disk(mddev, new_decode_dev(arg));
6957 /* We can support ADD_NEW_DISK on read-only arrays
6958 * only if we are re-adding a preexisting device.
6959 * So require mddev->pers and MD_DISK_SYNC.
6962 mdu_disk_info_t info;
6963 if (copy_from_user(&info, argp, sizeof(info)))
6965 else if (!(info.state & (1<<MD_DISK_SYNC)))
6966 /* Need to clear read-only for this */
6969 err = add_new_disk(mddev, &info);
6975 if (get_user(ro, (int __user *)(arg))) {
6981 /* if the bdev is going readonly the value of mddev->ro
6982 * does not matter, no writes are coming
6987 /* are we are already prepared for writes? */
6991 /* transitioning to readauto need only happen for
6992 * arrays that call md_write_start
6995 err = restart_array(mddev);
6998 set_disk_ro(mddev->gendisk, 0);
7005 * The remaining ioctls are changing the state of the
7006 * superblock, so we do not allow them on read-only arrays.
7008 if (mddev->ro && mddev->pers) {
7009 if (mddev->ro == 2) {
7011 sysfs_notify_dirent_safe(mddev->sysfs_state);
7012 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7013 /* mddev_unlock will wake thread */
7014 /* If a device failed while we were read-only, we
7015 * need to make sure the metadata is updated now.
7017 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
7018 mddev_unlock(mddev);
7019 wait_event(mddev->sb_wait,
7020 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
7021 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7022 mddev_lock_nointr(mddev);
7033 mdu_disk_info_t info;
7034 if (copy_from_user(&info, argp, sizeof(info)))
7037 err = add_new_disk(mddev, &info);
7041 case CLUSTERED_DISK_NACK:
7042 if (mddev_is_clustered(mddev))
7043 md_cluster_ops->new_disk_ack(mddev, false);
7049 err = hot_add_disk(mddev, new_decode_dev(arg));
7053 err = do_md_run(mddev);
7056 case SET_BITMAP_FILE:
7057 err = set_bitmap_file(mddev, (int)arg);
7066 if (mddev->hold_active == UNTIL_IOCTL &&
7068 mddev->hold_active = 0;
7069 mddev_unlock(mddev);
7071 if(did_set_md_closing)
7072 clear_bit(MD_CLOSING, &mddev->flags);
7075 #ifdef CONFIG_COMPAT
7076 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7077 unsigned int cmd, unsigned long arg)
7080 case HOT_REMOVE_DISK:
7082 case SET_DISK_FAULTY:
7083 case SET_BITMAP_FILE:
7084 /* These take in integer arg, do not convert */
7087 arg = (unsigned long)compat_ptr(arg);
7091 return md_ioctl(bdev, mode, cmd, arg);
7093 #endif /* CONFIG_COMPAT */
7095 static int md_open(struct block_device *bdev, fmode_t mode)
7098 * Succeed if we can lock the mddev, which confirms that
7099 * it isn't being stopped right now.
7101 struct mddev *mddev = mddev_find(bdev->bd_dev);
7107 if (mddev->gendisk != bdev->bd_disk) {
7108 /* we are racing with mddev_put which is discarding this
7112 /* Wait until bdev->bd_disk is definitely gone */
7113 if (work_pending(&mddev->del_work))
7114 flush_workqueue(md_misc_wq);
7117 BUG_ON(mddev != bdev->bd_disk->private_data);
7119 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7122 if (test_bit(MD_CLOSING, &mddev->flags)) {
7123 mutex_unlock(&mddev->open_mutex);
7129 atomic_inc(&mddev->openers);
7130 mutex_unlock(&mddev->open_mutex);
7132 check_disk_change(bdev);
7139 static void md_release(struct gendisk *disk, fmode_t mode)
7141 struct mddev *mddev = disk->private_data;
7144 atomic_dec(&mddev->openers);
7148 static int md_media_changed(struct gendisk *disk)
7150 struct mddev *mddev = disk->private_data;
7152 return mddev->changed;
7155 static int md_revalidate(struct gendisk *disk)
7157 struct mddev *mddev = disk->private_data;
7162 static const struct block_device_operations md_fops =
7164 .owner = THIS_MODULE,
7166 .release = md_release,
7168 #ifdef CONFIG_COMPAT
7169 .compat_ioctl = md_compat_ioctl,
7171 .getgeo = md_getgeo,
7172 .media_changed = md_media_changed,
7173 .revalidate_disk= md_revalidate,
7176 static int md_thread(void *arg)
7178 struct md_thread *thread = arg;
7181 * md_thread is a 'system-thread', it's priority should be very
7182 * high. We avoid resource deadlocks individually in each
7183 * raid personality. (RAID5 does preallocation) We also use RR and
7184 * the very same RT priority as kswapd, thus we will never get
7185 * into a priority inversion deadlock.
7187 * we definitely have to have equal or higher priority than
7188 * bdflush, otherwise bdflush will deadlock if there are too
7189 * many dirty RAID5 blocks.
7192 allow_signal(SIGKILL);
7193 while (!kthread_should_stop()) {
7195 /* We need to wait INTERRUPTIBLE so that
7196 * we don't add to the load-average.
7197 * That means we need to be sure no signals are
7200 if (signal_pending(current))
7201 flush_signals(current);
7203 wait_event_interruptible_timeout
7205 test_bit(THREAD_WAKEUP, &thread->flags)
7206 || kthread_should_stop(),
7209 clear_bit(THREAD_WAKEUP, &thread->flags);
7210 if (!kthread_should_stop())
7211 thread->run(thread);
7217 void md_wakeup_thread(struct md_thread *thread)
7220 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7221 set_bit(THREAD_WAKEUP, &thread->flags);
7222 wake_up(&thread->wqueue);
7225 EXPORT_SYMBOL(md_wakeup_thread);
7227 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7228 struct mddev *mddev, const char *name)
7230 struct md_thread *thread;
7232 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7236 init_waitqueue_head(&thread->wqueue);
7239 thread->mddev = mddev;
7240 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7241 thread->tsk = kthread_run(md_thread, thread,
7243 mdname(thread->mddev),
7245 if (IS_ERR(thread->tsk)) {
7251 EXPORT_SYMBOL(md_register_thread);
7253 void md_unregister_thread(struct md_thread **threadp)
7255 struct md_thread *thread = *threadp;
7258 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7259 /* Locking ensures that mddev_unlock does not wake_up a
7260 * non-existent thread
7262 spin_lock(&pers_lock);
7264 spin_unlock(&pers_lock);
7266 kthread_stop(thread->tsk);
7269 EXPORT_SYMBOL(md_unregister_thread);
7271 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7273 if (!rdev || test_bit(Faulty, &rdev->flags))
7276 if (!mddev->pers || !mddev->pers->error_handler)
7278 mddev->pers->error_handler(mddev,rdev);
7279 if (mddev->degraded)
7280 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7281 sysfs_notify_dirent_safe(rdev->sysfs_state);
7282 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7283 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7284 md_wakeup_thread(mddev->thread);
7285 if (mddev->event_work.func)
7286 queue_work(md_misc_wq, &mddev->event_work);
7287 md_new_event(mddev);
7289 EXPORT_SYMBOL(md_error);
7291 /* seq_file implementation /proc/mdstat */
7293 static void status_unused(struct seq_file *seq)
7296 struct md_rdev *rdev;
7298 seq_printf(seq, "unused devices: ");
7300 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7301 char b[BDEVNAME_SIZE];
7303 seq_printf(seq, "%s ",
7304 bdevname(rdev->bdev,b));
7307 seq_printf(seq, "<none>");
7309 seq_printf(seq, "\n");
7312 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7314 sector_t max_sectors, resync, res;
7315 unsigned long dt, db = 0;
7316 sector_t rt, curr_mark_cnt, resync_mark_cnt;
7317 int scale, recovery_active;
7318 unsigned int per_milli;
7320 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7321 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7322 max_sectors = mddev->resync_max_sectors;
7324 max_sectors = mddev->dev_sectors;
7326 resync = mddev->curr_resync;
7328 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7329 /* Still cleaning up */
7330 resync = max_sectors;
7332 resync -= atomic_read(&mddev->recovery_active);
7335 if (mddev->recovery_cp < MaxSector) {
7336 seq_printf(seq, "\tresync=PENDING");
7342 seq_printf(seq, "\tresync=DELAYED");
7346 WARN_ON(max_sectors == 0);
7347 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7348 * in a sector_t, and (max_sectors>>scale) will fit in a
7349 * u32, as those are the requirements for sector_div.
7350 * Thus 'scale' must be at least 10
7353 if (sizeof(sector_t) > sizeof(unsigned long)) {
7354 while ( max_sectors/2 > (1ULL<<(scale+32)))
7357 res = (resync>>scale)*1000;
7358 sector_div(res, (u32)((max_sectors>>scale)+1));
7362 int i, x = per_milli/50, y = 20-x;
7363 seq_printf(seq, "[");
7364 for (i = 0; i < x; i++)
7365 seq_printf(seq, "=");
7366 seq_printf(seq, ">");
7367 for (i = 0; i < y; i++)
7368 seq_printf(seq, ".");
7369 seq_printf(seq, "] ");
7371 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7372 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7374 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7376 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7377 "resync" : "recovery"))),
7378 per_milli/10, per_milli % 10,
7379 (unsigned long long) resync/2,
7380 (unsigned long long) max_sectors/2);
7383 * dt: time from mark until now
7384 * db: blocks written from mark until now
7385 * rt: remaining time
7387 * rt is a sector_t, which is always 64bit now. We are keeping
7388 * the original algorithm, but it is not really necessary.
7390 * Original algorithm:
7391 * So we divide before multiply in case it is 32bit and close
7393 * We scale the divisor (db) by 32 to avoid losing precision
7394 * near the end of resync when the number of remaining sectors
7396 * We then divide rt by 32 after multiplying by db to compensate.
7397 * The '+1' avoids division by zero if db is very small.
7399 dt = ((jiffies - mddev->resync_mark) / HZ);
7402 curr_mark_cnt = mddev->curr_mark_cnt;
7403 recovery_active = atomic_read(&mddev->recovery_active);
7404 resync_mark_cnt = mddev->resync_mark_cnt;
7406 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
7407 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
7409 rt = max_sectors - resync; /* number of remaining sectors */
7410 rt = div64_u64(rt, db/32+1);
7414 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7415 ((unsigned long)rt % 60)/6);
7417 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7421 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7423 struct list_head *tmp;
7425 struct mddev *mddev;
7433 spin_lock(&all_mddevs_lock);
7434 list_for_each(tmp,&all_mddevs)
7436 mddev = list_entry(tmp, struct mddev, all_mddevs);
7438 spin_unlock(&all_mddevs_lock);
7441 spin_unlock(&all_mddevs_lock);
7443 return (void*)2;/* tail */
7447 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7449 struct list_head *tmp;
7450 struct mddev *next_mddev, *mddev = v;
7456 spin_lock(&all_mddevs_lock);
7458 tmp = all_mddevs.next;
7460 tmp = mddev->all_mddevs.next;
7461 if (tmp != &all_mddevs)
7462 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7464 next_mddev = (void*)2;
7467 spin_unlock(&all_mddevs_lock);
7475 static void md_seq_stop(struct seq_file *seq, void *v)
7477 struct mddev *mddev = v;
7479 if (mddev && v != (void*)1 && v != (void*)2)
7483 static int md_seq_show(struct seq_file *seq, void *v)
7485 struct mddev *mddev = v;
7487 struct md_rdev *rdev;
7489 if (v == (void*)1) {
7490 struct md_personality *pers;
7491 seq_printf(seq, "Personalities : ");
7492 spin_lock(&pers_lock);
7493 list_for_each_entry(pers, &pers_list, list)
7494 seq_printf(seq, "[%s] ", pers->name);
7496 spin_unlock(&pers_lock);
7497 seq_printf(seq, "\n");
7498 seq->poll_event = atomic_read(&md_event_count);
7501 if (v == (void*)2) {
7506 spin_lock(&mddev->lock);
7507 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7508 seq_printf(seq, "%s : %sactive", mdname(mddev),
7509 mddev->pers ? "" : "in");
7512 seq_printf(seq, " (read-only)");
7514 seq_printf(seq, " (auto-read-only)");
7515 seq_printf(seq, " %s", mddev->pers->name);
7520 rdev_for_each_rcu(rdev, mddev) {
7521 char b[BDEVNAME_SIZE];
7522 seq_printf(seq, " %s[%d]",
7523 bdevname(rdev->bdev,b), rdev->desc_nr);
7524 if (test_bit(WriteMostly, &rdev->flags))
7525 seq_printf(seq, "(W)");
7526 if (test_bit(Journal, &rdev->flags))
7527 seq_printf(seq, "(J)");
7528 if (test_bit(Faulty, &rdev->flags)) {
7529 seq_printf(seq, "(F)");
7532 if (rdev->raid_disk < 0)
7533 seq_printf(seq, "(S)"); /* spare */
7534 if (test_bit(Replacement, &rdev->flags))
7535 seq_printf(seq, "(R)");
7536 sectors += rdev->sectors;
7540 if (!list_empty(&mddev->disks)) {
7542 seq_printf(seq, "\n %llu blocks",
7543 (unsigned long long)
7544 mddev->array_sectors / 2);
7546 seq_printf(seq, "\n %llu blocks",
7547 (unsigned long long)sectors / 2);
7549 if (mddev->persistent) {
7550 if (mddev->major_version != 0 ||
7551 mddev->minor_version != 90) {
7552 seq_printf(seq," super %d.%d",
7553 mddev->major_version,
7554 mddev->minor_version);
7556 } else if (mddev->external)
7557 seq_printf(seq, " super external:%s",
7558 mddev->metadata_type);
7560 seq_printf(seq, " super non-persistent");
7563 mddev->pers->status(seq, mddev);
7564 seq_printf(seq, "\n ");
7565 if (mddev->pers->sync_request) {
7566 if (status_resync(seq, mddev))
7567 seq_printf(seq, "\n ");
7570 seq_printf(seq, "\n ");
7572 bitmap_status(seq, mddev->bitmap);
7574 seq_printf(seq, "\n");
7576 spin_unlock(&mddev->lock);
7581 static const struct seq_operations md_seq_ops = {
7582 .start = md_seq_start,
7583 .next = md_seq_next,
7584 .stop = md_seq_stop,
7585 .show = md_seq_show,
7588 static int md_seq_open(struct inode *inode, struct file *file)
7590 struct seq_file *seq;
7593 error = seq_open(file, &md_seq_ops);
7597 seq = file->private_data;
7598 seq->poll_event = atomic_read(&md_event_count);
7602 static int md_unloading;
7603 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7605 struct seq_file *seq = filp->private_data;
7609 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7610 poll_wait(filp, &md_event_waiters, wait);
7612 /* always allow read */
7613 mask = POLLIN | POLLRDNORM;
7615 if (seq->poll_event != atomic_read(&md_event_count))
7616 mask |= POLLERR | POLLPRI;
7620 static const struct file_operations md_seq_fops = {
7621 .owner = THIS_MODULE,
7622 .open = md_seq_open,
7624 .llseek = seq_lseek,
7625 .release = seq_release_private,
7626 .poll = mdstat_poll,
7629 int register_md_personality(struct md_personality *p)
7631 printk(KERN_INFO "md: %s personality registered for level %d\n",
7633 spin_lock(&pers_lock);
7634 list_add_tail(&p->list, &pers_list);
7635 spin_unlock(&pers_lock);
7638 EXPORT_SYMBOL(register_md_personality);
7640 int unregister_md_personality(struct md_personality *p)
7642 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7643 spin_lock(&pers_lock);
7644 list_del_init(&p->list);
7645 spin_unlock(&pers_lock);
7648 EXPORT_SYMBOL(unregister_md_personality);
7650 int register_md_cluster_operations(struct md_cluster_operations *ops,
7651 struct module *module)
7654 spin_lock(&pers_lock);
7655 if (md_cluster_ops != NULL)
7658 md_cluster_ops = ops;
7659 md_cluster_mod = module;
7661 spin_unlock(&pers_lock);
7664 EXPORT_SYMBOL(register_md_cluster_operations);
7666 int unregister_md_cluster_operations(void)
7668 spin_lock(&pers_lock);
7669 md_cluster_ops = NULL;
7670 spin_unlock(&pers_lock);
7673 EXPORT_SYMBOL(unregister_md_cluster_operations);
7675 int md_setup_cluster(struct mddev *mddev, int nodes)
7677 if (!md_cluster_ops)
7678 request_module("md-cluster");
7679 spin_lock(&pers_lock);
7680 /* ensure module won't be unloaded */
7681 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7682 pr_err("can't find md-cluster module or get it's reference.\n");
7683 spin_unlock(&pers_lock);
7686 spin_unlock(&pers_lock);
7688 return md_cluster_ops->join(mddev, nodes);
7691 void md_cluster_stop(struct mddev *mddev)
7693 if (!md_cluster_ops)
7695 md_cluster_ops->leave(mddev);
7696 module_put(md_cluster_mod);
7699 static int is_mddev_idle(struct mddev *mddev, int init)
7701 struct md_rdev *rdev;
7707 rdev_for_each_rcu(rdev, mddev) {
7708 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7709 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7710 (int)part_stat_read(&disk->part0, sectors[1]) -
7711 atomic_read(&disk->sync_io);
7712 /* sync IO will cause sync_io to increase before the disk_stats
7713 * as sync_io is counted when a request starts, and
7714 * disk_stats is counted when it completes.
7715 * So resync activity will cause curr_events to be smaller than
7716 * when there was no such activity.
7717 * non-sync IO will cause disk_stat to increase without
7718 * increasing sync_io so curr_events will (eventually)
7719 * be larger than it was before. Once it becomes
7720 * substantially larger, the test below will cause
7721 * the array to appear non-idle, and resync will slow
7723 * If there is a lot of outstanding resync activity when
7724 * we set last_event to curr_events, then all that activity
7725 * completing might cause the array to appear non-idle
7726 * and resync will be slowed down even though there might
7727 * not have been non-resync activity. This will only
7728 * happen once though. 'last_events' will soon reflect
7729 * the state where there is little or no outstanding
7730 * resync requests, and further resync activity will
7731 * always make curr_events less than last_events.
7734 if (init || curr_events - rdev->last_events > 64) {
7735 rdev->last_events = curr_events;
7743 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7745 /* another "blocks" (512byte) blocks have been synced */
7746 atomic_sub(blocks, &mddev->recovery_active);
7747 wake_up(&mddev->recovery_wait);
7749 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7750 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7751 md_wakeup_thread(mddev->thread);
7752 // stop recovery, signal do_sync ....
7755 EXPORT_SYMBOL(md_done_sync);
7757 /* md_write_start(mddev, bi)
7758 * If we need to update some array metadata (e.g. 'active' flag
7759 * in superblock) before writing, schedule a superblock update
7760 * and wait for it to complete.
7762 void md_write_start(struct mddev *mddev, struct bio *bi)
7765 if (bio_data_dir(bi) != WRITE)
7768 BUG_ON(mddev->ro == 1);
7769 if (mddev->ro == 2) {
7770 /* need to switch to read/write */
7772 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7773 md_wakeup_thread(mddev->thread);
7774 md_wakeup_thread(mddev->sync_thread);
7777 atomic_inc(&mddev->writes_pending);
7778 if (mddev->safemode == 1)
7779 mddev->safemode = 0;
7780 if (mddev->in_sync) {
7781 spin_lock(&mddev->lock);
7782 if (mddev->in_sync) {
7784 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7785 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7786 md_wakeup_thread(mddev->thread);
7789 spin_unlock(&mddev->lock);
7792 sysfs_notify_dirent_safe(mddev->sysfs_state);
7793 wait_event(mddev->sb_wait,
7794 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7796 EXPORT_SYMBOL(md_write_start);
7798 void md_write_end(struct mddev *mddev)
7800 if (atomic_dec_and_test(&mddev->writes_pending)) {
7801 if (mddev->safemode == 2)
7802 md_wakeup_thread(mddev->thread);
7803 else if (mddev->safemode_delay)
7804 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7807 EXPORT_SYMBOL(md_write_end);
7809 /* md_allow_write(mddev)
7810 * Calling this ensures that the array is marked 'active' so that writes
7811 * may proceed without blocking. It is important to call this before
7812 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7813 * Must be called with mddev_lock held.
7815 * In the ->external case MD_CHANGE_PENDING can not be cleared until mddev->lock
7816 * is dropped, so return -EAGAIN after notifying userspace.
7818 int md_allow_write(struct mddev *mddev)
7824 if (!mddev->pers->sync_request)
7827 spin_lock(&mddev->lock);
7828 if (mddev->in_sync) {
7830 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7831 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7832 if (mddev->safemode_delay &&
7833 mddev->safemode == 0)
7834 mddev->safemode = 1;
7835 spin_unlock(&mddev->lock);
7836 md_update_sb(mddev, 0);
7837 sysfs_notify_dirent_safe(mddev->sysfs_state);
7839 spin_unlock(&mddev->lock);
7841 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7846 EXPORT_SYMBOL_GPL(md_allow_write);
7848 #define SYNC_MARKS 10
7849 #define SYNC_MARK_STEP (3*HZ)
7850 #define UPDATE_FREQUENCY (5*60*HZ)
7851 void md_do_sync(struct md_thread *thread)
7853 struct mddev *mddev = thread->mddev;
7854 struct mddev *mddev2;
7855 unsigned int currspeed = 0,
7857 sector_t max_sectors,j, io_sectors, recovery_done;
7858 unsigned long mark[SYNC_MARKS];
7859 unsigned long update_time;
7860 sector_t mark_cnt[SYNC_MARKS];
7862 struct list_head *tmp;
7863 sector_t last_check;
7865 struct md_rdev *rdev;
7866 char *desc, *action = NULL;
7867 struct blk_plug plug;
7870 /* just incase thread restarts... */
7871 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7873 if (mddev->ro) {/* never try to sync a read-only array */
7874 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7878 if (mddev_is_clustered(mddev)) {
7879 ret = md_cluster_ops->resync_start(mddev);
7883 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
7884 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7885 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
7886 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
7887 && ((unsigned long long)mddev->curr_resync_completed
7888 < (unsigned long long)mddev->resync_max_sectors))
7892 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7893 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7894 desc = "data-check";
7896 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7897 desc = "requested-resync";
7901 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7906 mddev->last_sync_action = action ?: desc;
7908 /* we overload curr_resync somewhat here.
7909 * 0 == not engaged in resync at all
7910 * 2 == checking that there is no conflict with another sync
7911 * 1 == like 2, but have yielded to allow conflicting resync to
7913 * other == active in resync - this many blocks
7915 * Before starting a resync we must have set curr_resync to
7916 * 2, and then checked that every "conflicting" array has curr_resync
7917 * less than ours. When we find one that is the same or higher
7918 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7919 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7920 * This will mean we have to start checking from the beginning again.
7925 int mddev2_minor = -1;
7926 mddev->curr_resync = 2;
7929 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7931 for_each_mddev(mddev2, tmp) {
7932 if (mddev2 == mddev)
7934 if (!mddev->parallel_resync
7935 && mddev2->curr_resync
7936 && match_mddev_units(mddev, mddev2)) {
7938 if (mddev < mddev2 && mddev->curr_resync == 2) {
7939 /* arbitrarily yield */
7940 mddev->curr_resync = 1;
7941 wake_up(&resync_wait);
7943 if (mddev > mddev2 && mddev->curr_resync == 1)
7944 /* no need to wait here, we can wait the next
7945 * time 'round when curr_resync == 2
7948 /* We need to wait 'interruptible' so as not to
7949 * contribute to the load average, and not to
7950 * be caught by 'softlockup'
7952 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7953 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7954 mddev2->curr_resync >= mddev->curr_resync) {
7955 if (mddev2_minor != mddev2->md_minor) {
7956 mddev2_minor = mddev2->md_minor;
7957 printk(KERN_INFO "md: delaying %s of %s"
7958 " until %s has finished (they"
7959 " share one or more physical units)\n",
7960 desc, mdname(mddev),
7964 if (signal_pending(current))
7965 flush_signals(current);
7967 finish_wait(&resync_wait, &wq);
7970 finish_wait(&resync_wait, &wq);
7973 } while (mddev->curr_resync < 2);
7976 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7977 /* resync follows the size requested by the personality,
7978 * which defaults to physical size, but can be virtual size
7980 max_sectors = mddev->resync_max_sectors;
7981 atomic64_set(&mddev->resync_mismatches, 0);
7982 /* we don't use the checkpoint if there's a bitmap */
7983 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7984 j = mddev->resync_min;
7985 else if (!mddev->bitmap)
7986 j = mddev->recovery_cp;
7988 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7989 max_sectors = mddev->resync_max_sectors;
7991 /* recovery follows the physical size of devices */
7992 max_sectors = mddev->dev_sectors;
7995 rdev_for_each_rcu(rdev, mddev)
7996 if (rdev->raid_disk >= 0 &&
7997 !test_bit(Journal, &rdev->flags) &&
7998 !test_bit(Faulty, &rdev->flags) &&
7999 !test_bit(In_sync, &rdev->flags) &&
8000 rdev->recovery_offset < j)
8001 j = rdev->recovery_offset;
8004 /* If there is a bitmap, we need to make sure all
8005 * writes that started before we added a spare
8006 * complete before we start doing a recovery.
8007 * Otherwise the write might complete and (via
8008 * bitmap_endwrite) set a bit in the bitmap after the
8009 * recovery has checked that bit and skipped that
8012 if (mddev->bitmap) {
8013 mddev->pers->quiesce(mddev, 1);
8014 mddev->pers->quiesce(mddev, 0);
8018 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
8019 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
8020 " %d KB/sec/disk.\n", speed_min(mddev));
8021 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
8022 "(but not more than %d KB/sec) for %s.\n",
8023 speed_max(mddev), desc);
8025 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8028 for (m = 0; m < SYNC_MARKS; m++) {
8030 mark_cnt[m] = io_sectors;
8033 mddev->resync_mark = mark[last_mark];
8034 mddev->resync_mark_cnt = mark_cnt[last_mark];
8037 * Tune reconstruction:
8039 window = 32*(PAGE_SIZE/512);
8040 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
8041 window/2, (unsigned long long)max_sectors/2);
8043 atomic_set(&mddev->recovery_active, 0);
8048 "md: resuming %s of %s from checkpoint.\n",
8049 desc, mdname(mddev));
8050 mddev->curr_resync = j;
8052 mddev->curr_resync = 3; /* no longer delayed */
8053 mddev->curr_resync_completed = j;
8054 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8055 md_new_event(mddev);
8056 update_time = jiffies;
8058 blk_start_plug(&plug);
8059 while (j < max_sectors) {
8064 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8065 ((mddev->curr_resync > mddev->curr_resync_completed &&
8066 (mddev->curr_resync - mddev->curr_resync_completed)
8067 > (max_sectors >> 4)) ||
8068 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8069 (j - mddev->curr_resync_completed)*2
8070 >= mddev->resync_max - mddev->curr_resync_completed ||
8071 mddev->curr_resync_completed > mddev->resync_max
8073 /* time to update curr_resync_completed */
8074 wait_event(mddev->recovery_wait,
8075 atomic_read(&mddev->recovery_active) == 0);
8076 mddev->curr_resync_completed = j;
8077 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8078 j > mddev->recovery_cp)
8079 mddev->recovery_cp = j;
8080 update_time = jiffies;
8081 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8082 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8085 while (j >= mddev->resync_max &&
8086 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8087 /* As this condition is controlled by user-space,
8088 * we can block indefinitely, so use '_interruptible'
8089 * to avoid triggering warnings.
8091 flush_signals(current); /* just in case */
8092 wait_event_interruptible(mddev->recovery_wait,
8093 mddev->resync_max > j
8094 || test_bit(MD_RECOVERY_INTR,
8098 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8101 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8103 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8107 if (!skipped) { /* actual IO requested */
8108 io_sectors += sectors;
8109 atomic_add(sectors, &mddev->recovery_active);
8112 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8116 if (j > max_sectors)
8117 /* when skipping, extra large numbers can be returned. */
8120 mddev->curr_resync = j;
8121 mddev->curr_mark_cnt = io_sectors;
8122 if (last_check == 0)
8123 /* this is the earliest that rebuild will be
8124 * visible in /proc/mdstat
8126 md_new_event(mddev);
8128 if (last_check + window > io_sectors || j == max_sectors)
8131 last_check = io_sectors;
8133 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8135 int next = (last_mark+1) % SYNC_MARKS;
8137 mddev->resync_mark = mark[next];
8138 mddev->resync_mark_cnt = mark_cnt[next];
8139 mark[next] = jiffies;
8140 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8144 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8148 * this loop exits only if either when we are slower than
8149 * the 'hard' speed limit, or the system was IO-idle for
8151 * the system might be non-idle CPU-wise, but we only care
8152 * about not overloading the IO subsystem. (things like an
8153 * e2fsck being done on the RAID array should execute fast)
8157 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8158 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8159 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8161 if (currspeed > speed_min(mddev)) {
8162 if (currspeed > speed_max(mddev)) {
8166 if (!is_mddev_idle(mddev, 0)) {
8168 * Give other IO more of a chance.
8169 * The faster the devices, the less we wait.
8171 wait_event(mddev->recovery_wait,
8172 !atomic_read(&mddev->recovery_active));
8176 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8177 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8178 ? "interrupted" : "done");
8180 * this also signals 'finished resyncing' to md_stop
8182 blk_finish_plug(&plug);
8183 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8185 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8186 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8187 mddev->curr_resync > 3) {
8188 mddev->curr_resync_completed = mddev->curr_resync;
8189 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8191 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8193 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8194 mddev->curr_resync > 3) {
8195 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8196 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8197 if (mddev->curr_resync >= mddev->recovery_cp) {
8199 "md: checkpointing %s of %s.\n",
8200 desc, mdname(mddev));
8201 if (test_bit(MD_RECOVERY_ERROR,
8203 mddev->recovery_cp =
8204 mddev->curr_resync_completed;
8206 mddev->recovery_cp =
8210 mddev->recovery_cp = MaxSector;
8212 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8213 mddev->curr_resync = MaxSector;
8215 rdev_for_each_rcu(rdev, mddev)
8216 if (rdev->raid_disk >= 0 &&
8217 mddev->delta_disks >= 0 &&
8218 !test_bit(Journal, &rdev->flags) &&
8219 !test_bit(Faulty, &rdev->flags) &&
8220 !test_bit(In_sync, &rdev->flags) &&
8221 rdev->recovery_offset < mddev->curr_resync)
8222 rdev->recovery_offset = mddev->curr_resync;
8227 /* set CHANGE_PENDING here since maybe another update is needed,
8228 * so other nodes are informed. It should be harmless for normal
8230 set_mask_bits(&mddev->flags, 0,
8231 BIT(MD_CHANGE_PENDING) | BIT(MD_CHANGE_DEVS));
8233 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8234 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8235 mddev->delta_disks > 0 &&
8236 mddev->pers->finish_reshape &&
8237 mddev->pers->size &&
8239 mddev_lock_nointr(mddev);
8240 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8241 mddev_unlock(mddev);
8242 set_capacity(mddev->gendisk, mddev->array_sectors);
8243 revalidate_disk(mddev->gendisk);
8246 spin_lock(&mddev->lock);
8247 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8248 /* We completed so min/max setting can be forgotten if used. */
8249 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8250 mddev->resync_min = 0;
8251 mddev->resync_max = MaxSector;
8252 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8253 mddev->resync_min = mddev->curr_resync_completed;
8254 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8255 mddev->curr_resync = 0;
8256 spin_unlock(&mddev->lock);
8258 wake_up(&resync_wait);
8259 md_wakeup_thread(mddev->thread);
8262 EXPORT_SYMBOL_GPL(md_do_sync);
8264 static int remove_and_add_spares(struct mddev *mddev,
8265 struct md_rdev *this)
8267 struct md_rdev *rdev;
8270 bool remove_some = false;
8272 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8273 /* Mustn't remove devices when resync thread is running */
8276 rdev_for_each(rdev, mddev) {
8277 if ((this == NULL || rdev == this) &&
8278 rdev->raid_disk >= 0 &&
8279 !test_bit(Blocked, &rdev->flags) &&
8280 test_bit(Faulty, &rdev->flags) &&
8281 atomic_read(&rdev->nr_pending)==0) {
8282 /* Faulty non-Blocked devices with nr_pending == 0
8283 * never get nr_pending incremented,
8284 * never get Faulty cleared, and never get Blocked set.
8285 * So we can synchronize_rcu now rather than once per device
8288 set_bit(RemoveSynchronized, &rdev->flags);
8294 rdev_for_each(rdev, mddev) {
8295 if ((this == NULL || rdev == this) &&
8296 rdev->raid_disk >= 0 &&
8297 !test_bit(Blocked, &rdev->flags) &&
8298 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8299 (!test_bit(In_sync, &rdev->flags) &&
8300 !test_bit(Journal, &rdev->flags))) &&
8301 atomic_read(&rdev->nr_pending)==0)) {
8302 if (mddev->pers->hot_remove_disk(
8303 mddev, rdev) == 0) {
8304 sysfs_unlink_rdev(mddev, rdev);
8305 rdev->saved_raid_disk = rdev->raid_disk;
8306 rdev->raid_disk = -1;
8310 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8311 clear_bit(RemoveSynchronized, &rdev->flags);
8314 if (removed && mddev->kobj.sd)
8315 sysfs_notify(&mddev->kobj, NULL, "degraded");
8317 if (this && removed)
8320 rdev_for_each(rdev, mddev) {
8321 if (this && this != rdev)
8323 if (test_bit(Candidate, &rdev->flags))
8325 if (rdev->raid_disk >= 0 &&
8326 !test_bit(In_sync, &rdev->flags) &&
8327 !test_bit(Journal, &rdev->flags) &&
8328 !test_bit(Faulty, &rdev->flags))
8330 if (rdev->raid_disk >= 0)
8332 if (test_bit(Faulty, &rdev->flags))
8334 if (!test_bit(Journal, &rdev->flags)) {
8336 ! (rdev->saved_raid_disk >= 0 &&
8337 !test_bit(Bitmap_sync, &rdev->flags)))
8340 rdev->recovery_offset = 0;
8343 hot_add_disk(mddev, rdev) == 0) {
8344 if (sysfs_link_rdev(mddev, rdev))
8345 /* failure here is OK */;
8346 if (!test_bit(Journal, &rdev->flags))
8348 md_new_event(mddev);
8349 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8354 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8358 static void md_start_sync(struct work_struct *ws)
8360 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8362 mddev->sync_thread = md_register_thread(md_do_sync,
8365 if (!mddev->sync_thread) {
8366 printk(KERN_ERR "%s: could not start resync thread...\n",
8368 /* leave the spares where they are, it shouldn't hurt */
8369 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8370 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8371 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8372 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8373 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8374 wake_up(&resync_wait);
8375 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8377 if (mddev->sysfs_action)
8378 sysfs_notify_dirent_safe(mddev->sysfs_action);
8380 md_wakeup_thread(mddev->sync_thread);
8381 sysfs_notify_dirent_safe(mddev->sysfs_action);
8382 md_new_event(mddev);
8386 * This routine is regularly called by all per-raid-array threads to
8387 * deal with generic issues like resync and super-block update.
8388 * Raid personalities that don't have a thread (linear/raid0) do not
8389 * need this as they never do any recovery or update the superblock.
8391 * It does not do any resync itself, but rather "forks" off other threads
8392 * to do that as needed.
8393 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8394 * "->recovery" and create a thread at ->sync_thread.
8395 * When the thread finishes it sets MD_RECOVERY_DONE
8396 * and wakeups up this thread which will reap the thread and finish up.
8397 * This thread also removes any faulty devices (with nr_pending == 0).
8399 * The overall approach is:
8400 * 1/ if the superblock needs updating, update it.
8401 * 2/ If a recovery thread is running, don't do anything else.
8402 * 3/ If recovery has finished, clean up, possibly marking spares active.
8403 * 4/ If there are any faulty devices, remove them.
8404 * 5/ If array is degraded, try to add spares devices
8405 * 6/ If array has spares or is not in-sync, start a resync thread.
8407 void md_check_recovery(struct mddev *mddev)
8409 if (mddev->suspended)
8413 bitmap_daemon_work(mddev);
8415 if (signal_pending(current)) {
8416 if (mddev->pers->sync_request && !mddev->external) {
8417 printk(KERN_INFO "md: %s in immediate safe mode\n",
8419 mddev->safemode = 2;
8421 flush_signals(current);
8424 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8427 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8428 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8429 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8430 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8431 (mddev->external == 0 && mddev->safemode == 1) ||
8432 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8433 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8437 if (mddev_trylock(mddev)) {
8441 struct md_rdev *rdev;
8442 if (!mddev->external && mddev->in_sync)
8443 /* 'Blocked' flag not needed as failed devices
8444 * will be recorded if array switched to read/write.
8445 * Leaving it set will prevent the device
8446 * from being removed.
8448 rdev_for_each(rdev, mddev)
8449 clear_bit(Blocked, &rdev->flags);
8450 /* On a read-only array we can:
8451 * - remove failed devices
8452 * - add already-in_sync devices if the array itself
8454 * As we only add devices that are already in-sync,
8455 * we can activate the spares immediately.
8457 remove_and_add_spares(mddev, NULL);
8458 /* There is no thread, but we need to call
8459 * ->spare_active and clear saved_raid_disk
8461 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8462 md_reap_sync_thread(mddev);
8463 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8464 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8465 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8469 if (mddev_is_clustered(mddev)) {
8470 struct md_rdev *rdev, *tmp;
8471 /* kick the device if another node issued a
8474 rdev_for_each_safe(rdev, tmp, mddev) {
8475 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8476 rdev->raid_disk < 0)
8477 md_kick_rdev_from_array(rdev);
8480 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8481 md_reload_sb(mddev, mddev->good_device_nr);
8484 if (!mddev->external) {
8486 spin_lock(&mddev->lock);
8487 if (mddev->safemode &&
8488 !atomic_read(&mddev->writes_pending) &&
8490 mddev->recovery_cp == MaxSector) {
8493 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8495 if (mddev->safemode == 1)
8496 mddev->safemode = 0;
8497 spin_unlock(&mddev->lock);
8499 sysfs_notify_dirent_safe(mddev->sysfs_state);
8502 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8503 md_update_sb(mddev, 0);
8505 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8506 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8507 /* resync/recovery still happening */
8508 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8511 if (mddev->sync_thread) {
8512 md_reap_sync_thread(mddev);
8515 /* Set RUNNING before clearing NEEDED to avoid
8516 * any transients in the value of "sync_action".
8518 mddev->curr_resync_completed = 0;
8519 spin_lock(&mddev->lock);
8520 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8521 spin_unlock(&mddev->lock);
8522 /* Clear some bits that don't mean anything, but
8525 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8526 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8528 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8529 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8531 /* no recovery is running.
8532 * remove any failed drives, then
8533 * add spares if possible.
8534 * Spares are also removed and re-added, to allow
8535 * the personality to fail the re-add.
8538 if (mddev->reshape_position != MaxSector) {
8539 if (mddev->pers->check_reshape == NULL ||
8540 mddev->pers->check_reshape(mddev) != 0)
8541 /* Cannot proceed */
8543 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8544 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8545 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8546 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8547 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8548 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8549 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8550 } else if (mddev->recovery_cp < MaxSector) {
8551 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8552 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8553 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8554 /* nothing to be done ... */
8557 if (mddev->pers->sync_request) {
8559 /* We are adding a device or devices to an array
8560 * which has the bitmap stored on all devices.
8561 * So make sure all bitmap pages get written
8563 bitmap_write_all(mddev->bitmap);
8565 INIT_WORK(&mddev->del_work, md_start_sync);
8566 queue_work(md_misc_wq, &mddev->del_work);
8570 if (!mddev->sync_thread) {
8571 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8572 wake_up(&resync_wait);
8573 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8575 if (mddev->sysfs_action)
8576 sysfs_notify_dirent_safe(mddev->sysfs_action);
8579 wake_up(&mddev->sb_wait);
8580 mddev_unlock(mddev);
8583 EXPORT_SYMBOL(md_check_recovery);
8585 void md_reap_sync_thread(struct mddev *mddev)
8587 struct md_rdev *rdev;
8589 /* resync has finished, collect result */
8590 md_unregister_thread(&mddev->sync_thread);
8591 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8592 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
8593 mddev->degraded != mddev->raid_disks) {
8595 /* activate any spares */
8596 if (mddev->pers->spare_active(mddev)) {
8597 sysfs_notify(&mddev->kobj, NULL,
8599 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8602 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8603 mddev->pers->finish_reshape)
8604 mddev->pers->finish_reshape(mddev);
8606 /* If array is no-longer degraded, then any saved_raid_disk
8607 * information must be scrapped.
8609 if (!mddev->degraded)
8610 rdev_for_each(rdev, mddev)
8611 rdev->saved_raid_disk = -1;
8613 md_update_sb(mddev, 1);
8614 /* MD_CHANGE_PENDING should be cleared by md_update_sb, so we can
8615 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8617 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8618 md_cluster_ops->resync_finish(mddev);
8619 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8620 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8621 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8622 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8623 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8624 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8625 wake_up(&resync_wait);
8626 /* flag recovery needed just to double check */
8627 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8628 sysfs_notify_dirent_safe(mddev->sysfs_action);
8629 md_new_event(mddev);
8630 if (mddev->event_work.func)
8631 queue_work(md_misc_wq, &mddev->event_work);
8633 EXPORT_SYMBOL(md_reap_sync_thread);
8635 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8637 sysfs_notify_dirent_safe(rdev->sysfs_state);
8638 wait_event_timeout(rdev->blocked_wait,
8639 !test_bit(Blocked, &rdev->flags) &&
8640 !test_bit(BlockedBadBlocks, &rdev->flags),
8641 msecs_to_jiffies(5000));
8642 rdev_dec_pending(rdev, mddev);
8644 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8646 void md_finish_reshape(struct mddev *mddev)
8648 /* called be personality module when reshape completes. */
8649 struct md_rdev *rdev;
8651 rdev_for_each(rdev, mddev) {
8652 if (rdev->data_offset > rdev->new_data_offset)
8653 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8655 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8656 rdev->data_offset = rdev->new_data_offset;
8659 EXPORT_SYMBOL(md_finish_reshape);
8661 /* Bad block management */
8663 /* Returns 1 on success, 0 on failure */
8664 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8667 struct mddev *mddev = rdev->mddev;
8670 s += rdev->new_data_offset;
8672 s += rdev->data_offset;
8673 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8675 /* Make sure they get written out promptly */
8676 sysfs_notify_dirent_safe(rdev->sysfs_state);
8677 set_mask_bits(&mddev->flags, 0,
8678 BIT(MD_CHANGE_CLEAN) | BIT(MD_CHANGE_PENDING));
8679 md_wakeup_thread(rdev->mddev->thread);
8684 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8686 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8690 s += rdev->new_data_offset;
8692 s += rdev->data_offset;
8693 return badblocks_clear(&rdev->badblocks,
8696 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8698 static int md_notify_reboot(struct notifier_block *this,
8699 unsigned long code, void *x)
8701 struct list_head *tmp;
8702 struct mddev *mddev;
8705 for_each_mddev(mddev, tmp) {
8706 if (mddev_trylock(mddev)) {
8708 __md_stop_writes(mddev);
8709 if (mddev->persistent)
8710 mddev->safemode = 2;
8711 mddev_unlock(mddev);
8716 * certain more exotic SCSI devices are known to be
8717 * volatile wrt too early system reboots. While the
8718 * right place to handle this issue is the given
8719 * driver, we do want to have a safe RAID driver ...
8727 static struct notifier_block md_notifier = {
8728 .notifier_call = md_notify_reboot,
8730 .priority = INT_MAX, /* before any real devices */
8733 static void md_geninit(void)
8735 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8737 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8740 static int __init md_init(void)
8744 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8748 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8752 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8755 if ((ret = register_blkdev(0, "mdp")) < 0)
8759 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8760 md_probe, NULL, NULL);
8761 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8762 md_probe, NULL, NULL);
8764 register_reboot_notifier(&md_notifier);
8765 raid_table_header = register_sysctl_table(raid_root_table);
8771 unregister_blkdev(MD_MAJOR, "md");
8773 destroy_workqueue(md_misc_wq);
8775 destroy_workqueue(md_wq);
8780 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8782 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8783 struct md_rdev *rdev2, *tmp;
8785 char b[BDEVNAME_SIZE];
8787 /* Check for change of roles in the active devices */
8788 rdev_for_each_safe(rdev2, tmp, mddev) {
8789 if (test_bit(Faulty, &rdev2->flags))
8792 /* Check if the roles changed */
8793 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8795 if (test_bit(Candidate, &rdev2->flags)) {
8796 if (role == 0xfffe) {
8797 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8798 md_kick_rdev_from_array(rdev2);
8802 clear_bit(Candidate, &rdev2->flags);
8805 if (role != rdev2->raid_disk) {
8807 if (rdev2->raid_disk == -1 && role != 0xffff) {
8808 rdev2->saved_raid_disk = role;
8809 ret = remove_and_add_spares(mddev, rdev2);
8810 pr_info("Activated spare: %s\n",
8811 bdevname(rdev2->bdev,b));
8812 /* wakeup mddev->thread here, so array could
8813 * perform resync with the new activated disk */
8814 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8815 md_wakeup_thread(mddev->thread);
8819 * We just want to do the minimum to mark the disk
8820 * as faulty. The recovery is performed by the
8821 * one who initiated the error.
8823 if ((role == 0xfffe) || (role == 0xfffd)) {
8824 md_error(mddev, rdev2);
8825 clear_bit(Blocked, &rdev2->flags);
8830 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8831 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8833 /* Finally set the event to be up to date */
8834 mddev->events = le64_to_cpu(sb->events);
8837 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8840 struct page *swapout = rdev->sb_page;
8841 struct mdp_superblock_1 *sb;
8843 /* Store the sb page of the rdev in the swapout temporary
8844 * variable in case we err in the future
8846 rdev->sb_page = NULL;
8847 alloc_disk_sb(rdev);
8848 ClearPageUptodate(rdev->sb_page);
8849 rdev->sb_loaded = 0;
8850 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
8853 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8854 __func__, __LINE__, rdev->desc_nr, err);
8855 put_page(rdev->sb_page);
8856 rdev->sb_page = swapout;
8857 rdev->sb_loaded = 1;
8861 sb = page_address(rdev->sb_page);
8862 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8866 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8867 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8869 /* The other node finished recovery, call spare_active to set
8870 * device In_sync and mddev->degraded
8872 if (rdev->recovery_offset == MaxSector &&
8873 !test_bit(In_sync, &rdev->flags) &&
8874 mddev->pers->spare_active(mddev))
8875 sysfs_notify(&mddev->kobj, NULL, "degraded");
8881 void md_reload_sb(struct mddev *mddev, int nr)
8883 struct md_rdev *rdev;
8887 rdev_for_each_rcu(rdev, mddev) {
8888 if (rdev->desc_nr == nr)
8892 if (!rdev || rdev->desc_nr != nr) {
8893 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8897 err = read_rdev(mddev, rdev);
8901 check_sb_changes(mddev, rdev);
8903 /* Read all rdev's to update recovery_offset */
8904 rdev_for_each_rcu(rdev, mddev)
8905 read_rdev(mddev, rdev);
8907 EXPORT_SYMBOL(md_reload_sb);
8912 * Searches all registered partitions for autorun RAID arrays
8916 static DEFINE_MUTEX(detected_devices_mutex);
8917 static LIST_HEAD(all_detected_devices);
8918 struct detected_devices_node {
8919 struct list_head list;
8923 void md_autodetect_dev(dev_t dev)
8925 struct detected_devices_node *node_detected_dev;
8927 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8928 if (node_detected_dev) {
8929 node_detected_dev->dev = dev;
8930 mutex_lock(&detected_devices_mutex);
8931 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8932 mutex_unlock(&detected_devices_mutex);
8934 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8935 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8939 static void autostart_arrays(int part)
8941 struct md_rdev *rdev;
8942 struct detected_devices_node *node_detected_dev;
8944 int i_scanned, i_passed;
8949 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8951 mutex_lock(&detected_devices_mutex);
8952 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8954 node_detected_dev = list_entry(all_detected_devices.next,
8955 struct detected_devices_node, list);
8956 list_del(&node_detected_dev->list);
8957 dev = node_detected_dev->dev;
8958 kfree(node_detected_dev);
8959 mutex_unlock(&detected_devices_mutex);
8960 rdev = md_import_device(dev,0, 90);
8961 mutex_lock(&detected_devices_mutex);
8965 if (test_bit(Faulty, &rdev->flags))
8968 set_bit(AutoDetected, &rdev->flags);
8969 list_add(&rdev->same_set, &pending_raid_disks);
8972 mutex_unlock(&detected_devices_mutex);
8974 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8975 i_scanned, i_passed);
8977 autorun_devices(part);
8980 #endif /* !MODULE */
8982 static __exit void md_exit(void)
8984 struct mddev *mddev;
8985 struct list_head *tmp;
8988 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8989 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8991 unregister_blkdev(MD_MAJOR,"md");
8992 unregister_blkdev(mdp_major, "mdp");
8993 unregister_reboot_notifier(&md_notifier);
8994 unregister_sysctl_table(raid_table_header);
8996 /* We cannot unload the modules while some process is
8997 * waiting for us in select() or poll() - wake them up
9000 while (waitqueue_active(&md_event_waiters)) {
9001 /* not safe to leave yet */
9002 wake_up(&md_event_waiters);
9006 remove_proc_entry("mdstat", NULL);
9008 for_each_mddev(mddev, tmp) {
9009 export_array(mddev);
9010 mddev->hold_active = 0;
9012 destroy_workqueue(md_misc_wq);
9013 destroy_workqueue(md_wq);
9016 subsys_initcall(md_init);
9017 module_exit(md_exit)
9019 static int get_ro(char *buffer, struct kernel_param *kp)
9021 return sprintf(buffer, "%d", start_readonly);
9023 static int set_ro(const char *val, struct kernel_param *kp)
9025 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9028 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9029 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9030 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9032 MODULE_LICENSE("GPL");
9033 MODULE_DESCRIPTION("MD RAID framework");
9035 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);